Optical fiber distribution cabinet automatic wiring path optimization method and device

Abstract

The present invention relates to the technical field of wiring path optimization, and discloses a method and device for optimizing the automatic wiring path of a fiber optic wiring cabinet. The method comprises: meshing the fiber optic wiring cabinet, establishing a finite element model of optical fiber physical stress, and performing stress analysis calculation to obtain optical fiber node stress distribution data; performing correlation analysis between node stress value and path length to obtain optical fiber stress and wiring path correlation data; inputting the optical fiber stress and wiring path correlation data into an encoder-decoder structure for deep feature extraction to obtain a target feature parameter set; performing wiring path prediction on the target feature parameter set to obtain wiring path prediction data; constructing a Kriging proxy model based on the wiring path prediction data, and performing stress distribution optimization and solution through a double forgetting factor genetic algorithm to obtain an optimal wiring path solution. The present invention effectively avoids the stress damage problem caused by traditional empirical wiring and realizes multi-objective optimization of the wiring path.

Description

Technical Field

[0001] The present invention relates to the technical field of wiring path optimization, and in particular to a method and device for optimizing the automatic wiring path of an optical fiber wiring cabinet. Background Art

[0002] With the rapid development of optical fiber communication networks, optical fiber distribution cabinets, as key communication infrastructure, have a direct impact on the reliability and stability of the network due to their wiring quality. Traditional optical fiber wiring methods mainly rely on manual experience for wiring, lacking scientific physical stress analysis and optimization methods, which leads to stress damage such as excessive bending, twisting and extrusion of optical fibers during wiring, affecting the quality of signal transmission.

[0003] At present, although there are some automated wiring systems, these systems mainly focus on the geometric constraints of path planning and rarely consider the physical stress characteristics of optical fibers. In practical applications, due to the complex physical characteristics of optical fibers and the nonlinear relationship between stress distribution and path length, traditional wiring methods based on empirical rules are difficult to effectively balance the optimization goals of optical fiber physical stress and wiring paths, often leading to problems such as local stress concentration or lengthy paths. In addition, most of the existing optical fiber wiring optimization methods use a single optimization algorithm and lack the ability to deeply learn and multi-objective optimize the optical fiber stress distribution. When dealing with complex wiring tasks in large-scale optical fiber wiring cabinets, the optimization efficiency is low and it is difficult to adapt to the requirements of dynamically adjusted working conditions. In particular, when considering the stress accumulation effect of optical fibers during long-term use, existing methods cannot effectively predict and optimize the stress distribution state of optical fibers, affecting the service life and maintenance efficiency of optical fiber wiring cabinets. Summary of the invention

[0004] The present invention provides a method and device for optimizing the automatic wiring path of an optical fiber distribution cabinet, which effectively avoids the stress damage problem caused by traditional empirical wiring and realizes multi-objective optimization of the wiring path.

[0005] In a first aspect, the present invention provides a method for optimizing an automated wiring path in an optical fiber wiring cabinet, the method comprising:

[0006] Meshing the optical fiber distribution cabinet, establishing an optical fiber physical stress finite element model, and performing stress analysis and calculation on the optical fiber physical stress finite element model to obtain optical fiber node stress distribution data;

[0007] Based on the optical fiber node stress distribution data, a correlation analysis is performed between the node stress value and the path length to obtain optical fiber stress and wiring path correlation data;

[0008] Inputting the optical fiber stress and wiring path correlation data into an encoder-decoder structure for deep feature extraction to obtain a target feature parameter set;

[0009] Performing wiring path prediction on the target characteristic parameter set to obtain wiring path prediction data;

[0010] A Kriging proxy model is constructed based on the wiring path prediction data, and a double forgetting factor genetic algorithm is used to optimize the stress distribution to obtain the optimal wiring path solution for the optical fiber wiring cabinet.

[0011] In a second aspect, the present invention provides an automatic wiring path optimization device for an optical fiber wiring cabinet, the automatic wiring path optimization device for an optical fiber wiring cabinet comprising:

[0012] Establishing a module for meshing the optical fiber distribution cabinet, establishing an optical fiber physical stress finite element model, and performing stress analysis and calculation on the optical fiber physical stress finite element model to obtain optical fiber node stress distribution data;

[0013] An analysis module, used to perform correlation analysis between node stress value and path length based on the optical fiber node stress distribution data, and obtain correlation data between optical fiber stress and wiring path;

[0014] An extraction module, used for inputting the optical fiber stress and wiring path correlation data into an encoder-decoder structure for deep feature extraction to obtain a target feature parameter set;

[0015] A prediction module, used to perform wiring path prediction on the target characteristic parameter set to obtain wiring path prediction data;

[0016] The solution module is used to construct a Kriging proxy model based on the wiring path prediction data, and to optimize the stress distribution by using a double forgetting factor genetic algorithm to obtain the optimal wiring path solution for the optical fiber wiring cabinet.

[0017] The third aspect of the present invention provides an automatic wiring path optimization device for a fiber optic wiring cabinet, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the automatic wiring path optimization device for a fiber optic wiring cabinet to perform the above-mentioned automatic wiring path optimization method for a fiber optic wiring cabinet.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for optimizing the automatic wiring path of an optical fiber wiring cabinet.

[0019] In the technical solution provided by the present invention, by establishing a finite element model of optical fiber physical stress, combined with explicit dynamics solution and stress analysis, the stress distribution data of optical fiber nodes is accurately obtained, and the stress damage problem caused by traditional empirical wiring is effectively avoided; an encoder-decoder structure is used for deep feature extraction, and stress features and path features are integrated to achieve accurate modeling of optical fiber wiring paths, overcoming the problem that traditional methods are insufficient in extracting nonlinear features; a wiring path prediction method based on a dual support vector regression structure significantly improves the prediction accuracy through the synergy of a main regressor and a residual regressor, introduces a Kriging proxy model and a dual forgetting factor genetic algorithm, and achieves multi-objective optimization of wiring paths, taking into account the optimization goals of minimizing stress and minimizing path length; through the dynamic adjustment mechanism of the time forgetting factor and the space forgetting factor, the global search capability of the algorithm is enhanced, and the local optimal solution is avoided; a hierarchical proxy model and an adaptive population adjustment strategy are used to significantly improve the computational efficiency of the optimization algorithm, and it has strong robustness and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 A schematic diagram of a process flow of an automated wiring path optimization method for an optical fiber wiring cabinet provided in an embodiment of the present application;

[0022] Figure 2 A schematic block diagram of the structure of an automatic wiring path optimization device for an optical fiber wiring cabinet provided in an embodiment of the present application;

[0023] Figure 3 A schematic block diagram of the structure of an automated wiring path optimization device for an optical fiber wiring cabinet provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0026] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0028] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0029] See also Figure 1 , Figure 1 A flow chart of a method for optimizing the automatic wiring path of an optical fiber wiring cabinet provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method for optimizing the automatic wiring path of the optical fiber distribution cabinet provided in the embodiment of the present application includes steps S100 to S600.

[0030] Step S100, meshing the optical fiber distribution cabinet, establishing an optical fiber physical stress finite element model, and performing stress analysis and calculation on the optical fiber physical stress finite element model to obtain optical fiber node stress distribution data;

[0031] It is understandable that the execution subject of the present invention may be an automatic wiring path optimization device for an optical fiber distribution cabinet, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0032] Specifically, the optical fiber distribution cabinet is geometrically modeled by 3D modeling software to construct a true-to-scale 3D structural model. In the modeling process, the complex structural characteristics inside the optical fiber distribution cabinet, such as wiring ducts, curved channels, connection areas, etc., are fully considered to ensure that the model can accurately reflect the geometric shape and physical characteristics of the actual equipment. After completing the 3D modeling, the hexahedral unit mesh is divided. Hexahedral unit division is an efficient and stable finite element mesh generation method. Its advantages are high calculation accuracy and good deformation control. It is suitable for complex geometric bodies such as optical fiber distribution cabinets that contain a large number of bends and detailed structures. In the division process, the mesh density is adjusted according to the characteristics of different regions of the structure. For areas with concentrated stress (such as bending points, fixed points, etc.), a higher mesh density is used to ensure the accuracy of stress calculation, while for areas with small stress changes, a coarser mesh can be used to save computing resources. After the mesh division is completed, mesh unit data containing a large number of units and nodes is obtained. Based on the mesh unit data, physical property parameters are assigned to the optical fiber material. The material properties of the optical fiber, such as elastic modulus, Poisson's ratio, density, yield strength and other key mechanical parameters, are added to the grid unit data to form an initial finite element model with physical significance. The assignment of material properties needs to be based on the actual material properties of the optical fiber to ensure that the model can truly reflect the force and deformation behavior of the optical fiber under mechanical action. Apply gravity load and fixed constraints to the initial finite element model to determine the constraint boundary conditions of the model. Gravity load is the main source of external force on the optical fiber under its own weight, while fixed constraints are used to simulate the boundary conditions of the optical fiber in the wiring cabinet, such as the fixed effect of the clamping point and the role of path constraints. The application of these boundary conditions needs to be accurately simulated in combination with the installation status of the actual wiring cabinet to ensure that the finite element model can reflect the real force environment. After the boundary conditions are determined, the optical fiber physical stress finite element model is constructed based on these conditions to obtain the finite element model to be analyzed. The model is imported into the explicit dynamics solver, and the node stress is solved by numerical calculation to obtain the initial node stress data. The explicit dynamics solver is a tool for efficiently handling dynamic problems and is suitable for stress analysis scenarios involving large deformations and complex boundary conditions. During the calculation process, the bending characteristics of the optical fiber are analyzed in detail. By extracting and processing the node stress data in the bending area, the local stress changes of the optical fiber in the bending state are calculated to form the bending stress data. Based on the bending stress data, the vibration characteristics of the optical fiber are analyzed to obtain the vibration characteristic data of the optical fiber in a dynamic environment. The vibration characteristic analysis considers the response characteristics of the optical fiber when subjected to external disturbances, such as natural frequency, vibration mode, and amplitude changes. These characteristics help to evaluate the long-term stability of the optical fiber. By combining the vibration characteristic data with the initial node stress data, the stress distribution of the optical fiber is corrected to reflect the comprehensive impact of the dynamic environment on the force on the optical fiber, and the final optical fiber node stress distribution data is obtained.

[0033] Step S200, performing correlation analysis between node stress value and path length based on optical fiber node stress distribution data to obtain optical fiber stress and wiring path correlation data;

[0034] Specifically, the stress distribution data of the optical fiber node is subjected to Fourier transform, and the stress change information in the time domain is converted into frequency domain information to obtain the spectrum characteristics of the stress change of the optical fiber node. In the process of Fourier transform, the stress data is decomposed into multiple frequency components, each of which corresponds to a different vibration mode or periodic characteristic. Through this process, the stress vibration law of the optical fiber node under different external conditions is understood to form stress spectrum data. The main frequency component is extracted from the stress spectrum data, and the main frequency component can reflect the main characteristics of the stress change. By extracting the main frequency component, representative main frequency stress characteristic data is obtained. A static wiring experiment is performed on the optical fiber distribution cabinet to collect the stress value and path length data of the optical fiber node under fixed wiring conditions to form a static data set. The static wiring experiment directly links the stress condition of the node with the path length by providing a stable initial state for the optical fiber, ensuring the accuracy and repeatability of the data. After the static data collection is completed, the optical fiber distribution cabinet is dynamically adjusted to simulate various changes that occur in actual operation, including temperature changes, mechanical vibrations, and external load disturbances, and collects the real-time stress data and path length change data of the node under dynamic conditions to form a dynamic data set. The dynamic data set supplements the deficiencies of the static data set, so that the model can not only capture the path stress relationship under static conditions, but also reflect the impact of dynamic changes on the path stress distribution. The static data set is merged with the dynamic data set to construct a unified polynomial fitting sample. The polynomial fitting sample is normalized, and the model deviation caused by the difference in numerical scale is eliminated by converting the data into a unified dimension range to obtain standardized training data. Based on the standardized training data, the coefficients of the polynomial fitting model are calculated using the least squares method. The least squares method minimizes the sum of squares of the error between the model prediction value and the actual observation value, ensuring that the fitting function can accurately describe the nonlinear relationship between the node stress and the path length, and obtains a polynomial fitting model, which can reflect the dominant characteristics of the optical fiber stress distribution and can efficiently perform stress path correlation calculations with low computational complexity. The main frequency stress characteristics are input into the polynomial fitting model for stress path correlation calculation to obtain the correlation data between the optical fiber stress and the wiring path.

[0035] Step S300, inputting the optical fiber stress and wiring path correlation data into the encoder-decoder structure for deep feature extraction to obtain a target feature parameter set;

[0036] Specifically, the optical fiber stress and wiring path correlation data are segmented to obtain stress feature data and path feature data. The stress feature data is input into the first encoder in the encoder-decoder structure. The first encoder consists of 5 convolutional layers, each of which uses a convolution kernel with a step size of 2 for convolution operation, and gradually compresses the spatial dimension of the stress feature data by extracting local features. Each convolution operation is followed by a batch normalization layer, which is used to normalize the features to reduce the internal covariate offset problem during model training and improve the convergence speed and stability. At the same time, each convolution layer is used in conjunction with the LeakyReLU activation function to avoid the "death" phenomenon of neurons by introducing a slope less than 0, enhance the nonlinear ability of feature extraction, and output stress compression features. The path feature data is input into the second encoder for feature compression. The structure and parameter settings of the second encoder are exactly the same as those of the first encoder, both of which contain 5 convolutional layers, each of which is followed by a batch normalization layer, and the activation function is also LeakyReLU. The compression process of path feature data can effectively extract the core information of path characteristics, including path length, complexity, and geometric features, and generate path compression features. The stress compression feature and path compression feature are fused to obtain fused feature data. The fused feature data is input into the decoder for feature reconstruction. The design of the decoder is relatively symmetrical with that of the encoder. It contains 5 deconvolution layers. Each deconvolution layer uses a deconvolution kernel with a step size of 2 to restore the original data characteristics by gradually restoring the spatial resolution of the compressed features. After each deconvolution operation, a batch normalization layer is connected to eliminate the gradient instability problem that occurs during training and improve the generalization ability of the model. In order to enhance the expressive power of feature reconstruction, a residual connection is added after each deconvolution layer to directly superimpose the input features on the output features, thereby slowing down the loss of feature information during the decoding process. Each deconvolution layer uses the ReLU activation function to forward propagate nonlinear features, so that the decoder can better learn and restore feature information. The reconstructed feature data is optimized and enhanced by introducing a spatial attention mechanism. The role of the spatial attention mechanism is to give higher weights to important features, while suppressing redundant or unimportant features, and improving the model's ability to focus on key features. In the process of spatial attention enhancement, the importance distribution of each position in the feature map is calculated to generate attention weights, and these weights are multiplied by the feature data point by point to form enhanced feature data. Multi-scale feature extraction operations are performed on the enhanced feature data. Through multi-scale convolution or pooling operations, the model can capture feature patterns at different scales, especially in scenarios dealing with complex paths and variable stress distributions. Multi-scale features can provide more comprehensive and detailed inputs for subsequent optimization processes. Through multi-scale feature extraction, a target feature parameter set is generated.

[0037] Step S400, performing wiring path prediction on the target feature parameter set to obtain wiring path prediction data;

[0038] Specifically, the target feature parameter set is input into the radial basis kernel function for nonlinear mapping. The radial basis kernel function is a kernel method that maps the feature space from the original low dimension to the high dimension, so that the nonlinear features can show a clearer structural relationship in the linear space. In the mapping process, the complex nonlinear relationship in the feature parameter set is unfolded to form the kernel feature data. Based on the kernel feature data, the penalty factor and precision parameter of the support vector regression machine are determined by the grid search method to obtain the optimized hyperparameters. The performance of the support vector regression machine depends largely on the selection of the penalty factor and the precision parameter. The penalty factor controls the tolerance of the model to the training error, while the precision parameter determines the sensitivity of the model to the deviation. In order to find the optimal hyperparameter combination, the grid search method systematically traverses the preset parameter range and evaluates different parameter combinations in combination with the cross-validation technique to select the optimized hyperparameters that can maximize the model accuracy. After the optimized hyperparameters are determined, a dual support vector regression structure is constructed based on these parameters. The structure includes a main regressor and a residual regressor, which aims to achieve higher prediction accuracy through staged modeling and error correction. As a first-level regression model, the main regressor is responsible for capturing the main nonlinear relationship between the kernel feature data and the wiring path. In the training of the main regressor, the ε-insensitive loss function is used. This function ignores the deviation between the predicted value and the actual value within a certain range by setting an insensitive interval, thereby improving the robustness of the model to noise. After the main regressor performs a first-level regression calculation on the kernel feature data, the main prediction result is output. The residual calculation is performed on the main prediction result and the original kernel feature data, and the error between the main prediction result and the true value is extracted to form the residual feature data. These residual data reflect the subtle patterns and local nonlinear features that the main regressor fails to fully capture. In order to improve the prediction accuracy, the residual feature data is input into the residual regressor for secondary regression calculation. The residual regressor adopts the Huber loss function, which has the characteristics of sensitivity of mean square error and robustness of absolute error, and can effectively handle outliers and noise in the residual. The residual prediction result is obtained by regression calculation of the residual feature. The main prediction result and the residual prediction result are weighted fused to complete the final prediction of the wiring path. The weights of weighted fusion are dynamically adjusted according to the performance of the model on the validation set to ensure that the fusion result can comprehensively reflect the global feature capture ability of the main regressor and the local refinement correction ability of the residual regressor. Through the above steps, the wiring path prediction data is obtained.

[0039] Step S500: construct a Kriging proxy model based on the wiring path prediction data, and perform stress distribution optimization solution through a double forgetting factor genetic algorithm to obtain an optimal wiring path solution for the optical fiber wiring cabinet.

[0040] Specifically, Latin hypercube sampling is performed on the distribution path prediction data to obtain the initial training sample data. Latin hypercube sampling can effectively cover the distribution range of design variables in multidimensional space and avoid samples being too concentrated or sparse. The initial training sample data is input into the Gaussian process regressor, and the hyperparameters of the model, including the length scale of the kernel function and the noise variance, are optimized by the maximum likelihood estimation method to construct a basic Kriging interpolation model. Sequence sampling optimization is performed based on the basic Kriging interpolation model. New enhanced sampling data are generated by enhanced sampling of the uncertainty region of the model, which can help the model better capture the complex change characteristics in the high-dimensional input space. Based on the enhanced sampling data, a hierarchical proxy model is constructed, which integrates global trends and local characteristics and can achieve efficient approximation in complex optical fiber distribution optimization problems. In the hierarchical proxy model, the global proxy model uses a second-order polynomial regression function as a trend term to capture the overall change trend of the distribution path parameters, while the local proxy model uses the moving Kriging method to finely model the key areas, thereby maintaining accuracy in the global range while taking into account local detail changes. After the hierarchical proxy model is constructed, the population initialization stage is entered. The wiring path parameters are encoded, and the path design variables are represented as individuals of the genetic algorithm in an appropriate way. The time forgetting factor and the space forgetting factor are set for each individual. The time forgetting factor is used to dynamically adjust the fitness impact of the population in different evolutionary generations, while the space forgetting factor is used to balance the fitness contribution of individuals in different regions, so as to find an appropriate balance between global search and local optimization. The mechanism of double forgetting factors effectively improves the global search ability and local convergence efficiency of the algorithm, forming a coded population with dynamic adjustment ability. After the coded population is input into the genetic algorithm, a crossover operation is performed to generate new offspring individuals by exchanging some gene fragments of the parent individuals, thereby ensuring the diversity of the population. The Gaussian mutation operator is used to perform mutation operations on the offspring individuals to introduce new genetic characteristics and prevent the algorithm from falling into the local optimum. Based on the crossover and mutation operations, the optimized population is generated to provide candidate solutions for subsequent target evaluation. The optimized population is evaluated by multiple objectives, with stress minimization and path length minimization as the optimization objectives. The optimized population is stratified by fast non-dominated sorting, and the crowding distance of each individual is calculated. The non-dominated sorting divides the population into different levels and sorts them according to the distribution of individuals in the target space, while the crowding distance is used to measure the degree of dispersion of individuals in the target space to ensure the diversity of the population in the target space. Based on the multi-objective evaluation, a set of high-quality candidate solutions is screened out. The candidate solution set is fuzzy comprehensively evaluated. By constructing a triangular membership function, the performance of each candidate solution is converted into a membership value, and the maximum membership principle is used to select the optimal wiring path solution for the optical fiber distribution cabinet.

[0041] The optimization variables of the hierarchical agent model are encoded with real numbers. The core of wiring path optimization is to express the starting coordinates, end coordinates and intermediate control point coordinates of the wiring in mathematical form and map them to the [0,1] interval to complete the generation of normalized path parameters. After obtaining the normalized path parameters, a uniformly distributed random number is used to generate multiple initial individuals in the [0,1] interval to construct the initial population. The randomness of the uniform distribution can ensure that the initial population covers the design space and prevent the population from falling into the local optimum too early. The time dimension fitness value is calculated for each individual in the initial population, thereby introducing the time forgetting factor. The time dimension fitness value is obtained by analyzing the historical performance of the individual in the evolution process, so as to evaluate the adaptability of the individual at different times. The role of the time forgetting factor is to dynamically adjust the historical influence weight of the individual, give higher weights to individuals with excellent recent performance, and gradually reduce the influence of the performance in the earlier stage, so as to guide the population to converge faster towards the direction of the high-quality solution. By introducing the time forgetting factor, the initial population is updated to the time-weighted individual. Based on the time-weighted individual, the spatial dimension fitness value is calculated, and the spatial forgetting factor is introduced. The spatial forgetting factor is set according to the relative position of individuals in the population. Its goal is to moderately weaken individuals in densely distributed areas, thereby encouraging exploration of spatial areas that have not been fully searched. The spatial forgetting factor can reduce the fitness value of individuals that are too close to other individuals in the group, and give higher fitness to those individuals distributed in sparse areas, thereby enhancing the global search ability of the algorithm. By combining the temporal forgetting factor and the spatial forgetting factor, each individual in the population is updated to a double forgetting factor individual. These individuals can dynamically adjust the fitness value in both time and space dimensions, reflecting the adaptability of the optimization algorithm. The double forgetting factor individuals are sorted according to the fitness value, and the top-ranked individuals are selected to form an elite population. The elite population represents the optimal solution set in the current population. In order to maintain the diversity of the population and avoid premature convergence of elite individuals, the elite population is adaptively adjusted in size. The size of the elite population is dynamically adjusted according to the search state during the optimization process. For example, the population size is increased in the early stage of the algorithm to ensure the breadth of the search, and the population size is gradually reduced in the later stage to speed up the convergence. The dynamic adjustment mechanism can achieve a balance between the search range and search accuracy between different optimization stages and generate a dynamic population. A population distribution model is constructed based on the dynamic population. The population distribution model reveals the search density and aggregation of the population in different areas by analyzing the distribution characteristics of individuals in the design space. According to the population distribution characteristics, the individuals in the population are clustered and grouped, and individuals with similar characteristics are divided into the same group. Local search is performed within each group. The local search process explores potential high-quality solutions by strengthening the fine adjustment of the solutions within the group. The local search results are combined with the overall information of the distribution model to finally generate an optimized coding population.

[0042] In the embodiment of the present invention, by establishing a finite element model of optical fiber physical stress, combined with explicit dynamics solution and stress analysis, the stress distribution data of optical fiber nodes is accurately obtained, and the stress damage problem caused by traditional empirical wiring is effectively avoided; an encoder-decoder structure is used for deep feature extraction, and stress features and path features are integrated to achieve accurate modeling of optical fiber wiring paths, overcoming the problem that traditional methods are insufficient in extracting nonlinear features; a wiring path prediction method based on a dual support vector regression structure significantly improves the prediction accuracy through the synergy of a main regressor and a residual regressor, introduces a Kriging proxy model and a dual forgetting factor genetic algorithm, and achieves multi-objective optimization of wiring paths, taking into account the optimization goals of minimizing stress and minimizing path length; through the dynamic adjustment mechanism of the time forgetting factor and the space forgetting factor, the global search capability of the algorithm is enhanced, and the local optimal solution is avoided; a hierarchical proxy model and an adaptive population adjustment strategy are used to significantly improve the computational efficiency of the optimization algorithm, and it has strong robustness and generalization ability.

[0043] In a specific embodiment, the process of executing step S100 may specifically include the following steps:

[0044] Perform three-dimensional modeling and hexahedral unit division on the optical fiber distribution cabinet to obtain grid unit data, and add optical fiber material property parameters to the grid unit data to obtain an initialized finite element model;

[0045] Applying gravity load and fixed constraints to the initialized finite element model to obtain constraint boundary conditions, and constructing a fiber physical stress finite element model based on the constraint boundary conditions to obtain a finite element model to be analyzed;

[0046] The finite element model to be analyzed is imported into the explicit dynamics solver to perform node stress calculation to obtain initial node stress data, and the initial node stress data is subjected to optical fiber bending stress analysis to obtain bending stress data;

[0047] The optical fiber vibration characteristic analysis is performed based on the bending stress data to obtain the vibration characteristic data, and the initial node stress data is corrected according to the vibration characteristic data to obtain the optical fiber node stress distribution data.

[0048] Specifically, the geometric structure of the wiring cabinet is 3D modeled and hexahedral unit division is completed. After the accurate 3D model is generated by computer-aided design tools, it is imported into the finite element analysis tool to divide the geometric model into hexahedral mesh units to form a discretized finite element model. After the mesh division is completed, the material property parameters of the optical fiber are assigned to each unit, including the elastic modulus. , Poisson's ratio ,density , forming an initial finite element model. These parameters determine the elastic deformation capacity and mass distribution of the optical fiber material and are used to simulate the real mechanical behavior. Apply gravity load to the initial finite element model and fixed constraints to define its constraint boundary conditions. The expression of gravity load is:

[0049] ;

[0050] in, is the density of the optical fiber, is the volume of the optical fiber, is the acceleration due to gravity. A fixed constraint is implemented by setting the displacement of a specific node to zero, e.g. , indicating that the three-dimensional displacement of the node is completely restricted. After the boundary conditions are applied, the optical fiber physical stress finite element model is formed, and the finite element model to be analyzed is obtained. The finite element model to be analyzed is imported into the explicit dynamics solver, and the node stress is calculated using the following motion equation:

[0051] ;

[0052] in, is the mass matrix, is the node acceleration vector, is the damping matrix, is the node velocity vector, is the stiffness matrix, is the node displacement vector, is the external force vector. By solving this equation, the displacement and stress of each node under the action of external force and boundary conditions are calculated to form the initial node stress data. Based on the initial node stress data, the bending stress of the optical fiber is calculated. The bending stress is calculated by the following formula:

[0053] ;

[0054] in, is the bending moment, is the distance from the neutral axis of the cross section to the outer edge of the cross section, is the section moment of inertia. For a circular cross-section optical fiber, the expression for the moment of inertia is:

[0055] ;

[0056] in, is the radius of the fiber cross section. , analyze the stress distribution of the optical fiber in the bending area and identify the stress concentration area. Perform vibration characteristic analysis based on bending stress data and calculate the natural frequency of the optical fiber and vibration modes. The natural frequency is determined by the following formula:

[0057] ;

[0058] in, is the equivalent stiffness, is the equivalent mass. The equivalent stiffness is obtained by the following relationship

[0059] ;

[0060] in, is the length of the optical fiber. The vibration mode is obtained by solving the eigenvalue problem, and the eigenvalue equation is:

[0061] ;

[0062] in, is the eigenvalue, is the vibration mode vector. The initial node stress data is corrected according to the vibration characteristic data. The dynamic stress correction is based on the following formula:

[0063] ;

[0064] in, is the static stress, is the dynamic stress component. The dynamic stress component is calculated by superimposing the results of vibration analysis. Through the above steps, the corrected optical fiber node stress distribution data is obtained, which reflects the comprehensive stress state of the optical fiber during operation.

[0065] In a specific embodiment, the process of executing step S200 may specifically include the following steps:

[0066] Perform Fourier transform on the stress distribution data of the optical fiber node to obtain stress spectrum data, and extract the main frequency component of the stress spectrum data to obtain the main frequency stress characteristics;

[0067] Perform static wiring on the optical fiber distribution cabinet, collect node stress values ​​and distribution path length data to obtain a static data set, and dynamically adjust the optical fiber distribution cabinet to collect real-time stress and path length change data to obtain a dynamic data set;

[0068] The static data set and the dynamic data set are merged to construct a polynomial fitting sample to obtain fitting training data, and the fitting training data is normalized to obtain standardized training data;

[0069] Based on the standardized training data, the least square method is used to calculate the polynomial coefficients to obtain the polynomial fitting model. The main frequency stress characteristics are input into the polynomial fitting model to perform stress path correlation calculation to obtain the optical fiber stress and wiring path correlation data.

[0070] Specifically, the stress data of the optical fiber node is subjected to Fourier transform. Fourier transform is a mathematical transformation that can convert signals in the time domain or space domain into frequency domain representation. Fourier transform helps the system analyze the frequency components of stress changes in the optical fiber during the wiring process and reveals the main periodic factors that affect the wiring path. The Fourier transform formula is:

[0071] ;

[0072] in, is the frequency The stress spectrum under is the original stress time series, is an imaginary unit, is a time variable It is frequency. Fourier transform converts time domain signal into frequency domain signal, so that the system can view the contribution of different frequency components to fiber stress. The main frequency component is extracted from the stress spectrum data to determine the main frequency component of fiber node stress. The main frequency component is extracted by analyzing the spectrum, identifying the peak in the spectrum, or determining the main frequency by calculating the power spectrum density. Common extraction methods are:

[0073] ;

[0074] in, Indicates the main frequency, is the power spectrum density of the spectrum. By extracting the main frequency component, the main characteristics of the optical fiber stress are obtained, reflecting the periodic change law of stress. Static wiring is performed after the main frequency feature extraction. Static wiring refers to designing the preliminary layout path of the optical fiber without considering the dynamic change factors. During the static wiring process, the node stress value and the wiring path length data are collected. The node stress value represents the stress distribution of the optical fiber at each wiring node, and the wiring path length represents the optical fiber path length from the starting point to the end point. According to the static wiring configuration, a static data set is obtained. After the static wiring is completed, dynamic adjustment is performed, that is, the optical fiber wiring cabinet is adjusted to adapt to the stress changes in actual operation. During the dynamic adjustment process, real-time stress and path length change data are collected, and these data constitute a dynamic data set. The dynamic data set can reflect the relationship between stress and path length during actual operation, and reveal the stress distribution of optical fiber wiring under different operating conditions. The static data set is merged with the dynamic data set to obtain a comprehensive data set. The static data provides a basic stress-path relationship, while the dynamic data corrects the assumptions of static wiring so that the model can adapt to actual conditions. The two data sets are combined to construct a polynomial fitting sample to obtain fitting training data. These training data are used to establish a mathematical model between stress and path length. The fitted training data is normalized. The data is scaled to a standard range and scaled to the [0,1] interval. The formula is as follows:

[0075] ;

[0076] in, is the original data, is the normalized data, and are the minimum and maximum values ​​of the data set, respectively. After completing data normalization, the polynomial coefficients are calculated using the least squares method. The least squares method is an optimization method that finds the optimal fitting function by minimizing the sum of squared errors. For polynomial fitting, it is assumed that the fiber stress With path length The relationship between is a polynomial function:

[0077] ;

[0078] in, are the coefficients of the polynomial to be found, is the path length, is the stress value corresponding to the path. The goal of the least squares method is to minimize the square error between the predicted stress and the actual stress, that is:

[0079] ;

[0080] By minimizing the sum of squared errors, the optimal coefficients of the polynomial are obtained. These coefficients are used to build a fitting model. The main frequency stress characteristics are input into the polynomial fitting model, and the stress-path correlation calculation is performed to obtain the correlation data between the optical fiber stress and the wiring path.

[0081] In a specific embodiment, the process of executing step S300 may specifically include the following steps:

[0082] Performing data segmentation on the optical fiber stress and wiring path correlation data to obtain stress characteristic data and path characteristic data;

[0083] The stress feature data is input into the first encoder in the encoder-decoder structure for feature compression. The first encoder contains 5 convolutional layers. Each convolutional layer uses a convolution kernel with a step size of 2 and a LeakyReLU activation function, and a batch normalization layer is connected after each convolutional layer to obtain stress compression features.

[0084] Inputting the path feature data into a second encoder in an encoder-decoder structure for feature compression, wherein the encoder structure and parameters of the second encoder are the same as those of the first encoder, to obtain a path compression feature;

[0085] The stress compression feature and the path compression feature are fused to obtain fused feature data, and the fused feature data is input into the decoder in the encoder-decoder structure for feature reconstruction. The decoder contains 5 deconvolution layers, each of which uses a deconvolution kernel with a step size of 2 and a ReLU activation function, and each deconvolution layer is followed by a batch normalization layer and a residual connection to obtain reconstructed feature data;

[0086] The reconstructed feature data is spatially enhanced to obtain enhanced feature data, and multi-scale feature extraction is performed on the enhanced feature data to obtain a target feature parameter set.

[0087] Specifically, the raw data is properly processed so that it can be input into the deep learning model for effective feature learning. For the given optical fiber stress and wiring path association data, the data is segmented to obtain stress feature data and path feature data. By dividing the data set into two parts, the stress and path are separated based on the physical properties of the data or predefined labels, and each is used as an independent input feature set. For example, the optical fiber stress feature data is a sequence of stress values ​​under different paths, while the path feature data contains geometric information such as path length and bending angle. This segmentation method helps the model to extract different types of information separately and avoid cross-interference of information. The stress feature data is input to the first encoder in the encoder-decoder structure for feature compression. The first encoder includes multiple convolutional layers, each of which is used to extract higher-level abstract features from the input features. For the encoder in this example, it is set to five convolutional layers, each of which uses a convolution kernel with a stride of 2, and a batch normalization layer is connected after each convolutional layer. The batch normalization layer is used to accelerate the training process and improve the stability of the model. Its function is to standardize the input features of each layer, reduce the internal covariance offset problem, and thus improve the training efficiency. The convolution kernel used in each convolution layer gradually extracts finer-grained spatial features when processing data. For example, the operation of the convolution kernel is described by the following formula:

[0088] ;

[0089] in, is the output of the convolutional layer, is the activation function, using the LeakyReLU activation function, its formula is:

[0090] ;

[0091] in, is a small constant that controls the slope of the negative part. The function of each convolution layer is to gradually reduce the dimension of the input data to extract the compressed representation of the stress feature and obtain the stress compression feature. The output of each convolution layer gradually reduces the size of the feature map through a convolution operation with a step size of 2, while increasing the depth, and gradually maps the high-dimensional stress data into a more compact feature representation. At the same time, the path feature data is input into the second encoder in the encoder-decoder structure, and the structure and parameters of the second encoder are the same as those of the first encoder. The second encoder also contains five convolution layers, and each convolution layer uses the same convolution kernel and activation function settings as the first encoder. After the path feature data undergoes this convolution operation, the path compression feature is obtained. This process is similar to the compression process of the stress feature, and the purpose is to extract the potential high-level patterns and laws in the path data. The stress compression feature and the path compression feature are feature fused to obtain the fused feature data. The fused feature data is input into the decoder part of the encoder-decoder structure for feature reconstruction. The decoder is composed of multiple deconvolution layers, and the function of each deconvolution layer is to gradually map the low-dimensional features back to a higher-dimensional space, thereby restoring the original features or approximate features. Each deconvolution layer uses a deconvolution kernel with a step size of 2 and a ReLU activation function. The calculation formula of the deconvolution layer is:

[0092] ;

[0093] in, is the output of the deconvolution layer, is the input data, is the filter, is the bias, is the ReLU activation function. The purpose of the deconvolution operation is to gradually restore low-dimensional data to a higher-dimensional feature space for fine feature reconstruction. Each deconvolution layer is connected to a batch normalization layer and a residual connection. The batch normalization layer is used to improve the training process, and the residual connection helps to alleviate the gradient vanishing problem during deep network training, so that information can be effectively propagated to the deep layer. The reconstructed feature data is spatially enhanced. The role of the spatial attention mechanism is to automatically focus on important areas and enhance the model's attention to key information. In the optimization problem of the fiber optic wiring cabinet, the stress and path in certain areas are more important to the optimization results. The spatial attention mechanism calculates the attention weight of each spatial position and weights the features, thereby improving the expression ability of the model. The calculation of spatial attention is described by the following formula:

[0094] ;

[0095] in, is the spatial attention map, is the Sigmoid activation function, and is the weight matrix, is the convolution operation, Represents the convolution operation. Through this mechanism, the model can automatically adjust the degree of attention to different feature areas. Multi-scale feature extraction is performed on the enhanced feature data to obtain the target feature parameter set. Features are observed from different scales or different resolutions to capture more information. By introducing a multi-scale strategy, the model can learn the complex relationship between stress and path at different scales, improving the robustness and expression ability of feature extraction. The target feature parameter set is obtained.

[0096] In a specific embodiment, the process of executing step S400 may specifically include the following steps:

[0097] Input the target feature parameter set into the radial basis kernel function for nonlinear mapping to obtain kernel feature data;

[0098] Based on the kernel feature data, the penalty factor and precision parameter of the support vector regression machine are determined by the grid search method to obtain the optimized hyperparameters;

[0099] Based on the optimization of hyperparameters, a dual support vector regression structure is constructed, where the dual support vector regression structure includes a main regressor and a residual regressor;

[0100] The kernel feature data is input into the main regressor for first-level regression calculation. The main regressor adopts the ε-insensitive loss function to obtain the main prediction result.

[0101] Perform residual calculation on the main prediction result and the kernel feature data to obtain residual feature data;

[0102] The residual feature data is input into the residual regressor for secondary regression calculation. The residual regressor adopts the Huber loss function to obtain the residual prediction result.

[0103] The main prediction results and the residual prediction results are weighted and fused to obtain the wiring path prediction data.

[0104] Specifically, the kernel feature data is obtained by performing nonlinear mapping of the target feature parameter set with the radial basis kernel function. The radial basis kernel function is a kernel function used in machine learning algorithms such as support vector machines (SVM), which can effectively handle nonlinear relationships. The role of the radial basis kernel function is to map the target feature parameter set from the original space to the high-dimensional feature space, thereby finding the distribution characteristics of the data in the high-dimensional space, so that the model can better capture the nonlinear relationship between the data. The formula of the radial basis kernel function is:

[0105] ;

[0106] in, and are two input vectors, is the Euclidean distance between the two input vectors, is the parameter of the kernel function, which controls the distribution range of the high-dimensional space. Through the kernel function, the input data will be mapped to a nonlinear high-dimensional space, so that the support vector regression model can perform regression modeling more effectively in this space. Based on the kernel feature data, the penalty factor ( and precision parameters Grid search is a hyperparameter optimization method that traverses the predefined hyperparameter space, trains each set of hyperparameter combinations, and selects the hyperparameter combination that makes the model perform best. In support vector regression, the penalty factor Controls the tolerance to error, the larger A value of means that the model pays more attention to reducing the error during training, while a smaller A higher value of leads to a stronger regularization effect. This determines the error tolerance range of the support vector machine. The value makes the model more tolerant to errors, thereby reducing the complexity of the model. Through the grid search method, the optimal and The SVR model has the best prediction performance. Based on the optimized hyperparameters, a dual support vector regression structure is constructed. The dual support vector regression structure includes a main regressor and a residual regressor. The main regressor performs a preliminary regression calculation on the target feature to obtain a rough prediction result of the wiring path. The regressor uses -Insensitive loss function measures the prediction error of the model. -The insensitive loss function is defined as:

[0107] ;

[0108] in, is the model's predicted value, is the true value, is the given tolerance error. When the difference between the predicted value and the true value exceeds When to calculate; and when the difference between the predicted value and the true value is less than When , the loss is zero. The main regressor optimizes the model through this loss function to minimize the error within a predetermined range. After obtaining the prediction results of the main regressor, the residuals of these prediction results and the kernel feature data are calculated. The residual refers to the difference between the prediction results of the main regressor and the actual results, and its formula is:

[0109] ;

[0110] in, is the residual, Is the real result, is the prediction result of the main regressor. The calculation of the residual enables the system to identify the part that the main regressor failed to predict well. This part of the residual contains nonlinear information that cannot be explained by the main regressor and needs to be further modeled and corrected by the residual regressor. The residual regressor uses the Huber loss function for secondary regression calculation. The Huber loss function is a loss function that combines the mean square error and absolute error characteristics and can effectively handle outliers. In the Huber loss function, when the prediction error is small (less than a threshold When the prediction error is large (greater than the threshold When , the absolute error is used. The form of the Huber loss function is:

[0111] ;

[0112] in, is a preset threshold. is the model prediction result, is the true value. The Huber loss function is relatively gentle in dealing with large errors, so it can improve the robustness of the model in the presence of outliers. The residual regressor regresses the residual through the Huber loss function to model the residual to correct the prediction error of the main regressor and obtain a more accurate residual prediction result. The main prediction result and the residual prediction result are weighted fused to obtain the wiring path prediction data. Combining the advantages of the main regressor and the residual regressor, a more accurate path prediction result is obtained. The formula for weighted fusion is:

[0113] ;

[0114] in, is the final prediction result. is the prediction result of the main regressor, is the prediction result of the residual regressor, is a weighting factor that controls the relative contribution of the two. The selection is done by cross-validation or other methods, and the value is usually between 0 and 1. When it is close to 1, the prediction result of the main regressor has a larger weight, and vice versa, the prediction result of the residual regressor has a larger weight.

[0115] In a specific embodiment, the process of executing step S500 may specifically include the following steps:

[0116] Perform Latin hypercube sampling on the wiring path prediction data to obtain initial training sample data;

[0117] The initial training sample data is input into the Gaussian process regressor, and the optimal parameter value is determined by the maximum likelihood estimation method to obtain the basic Kriging interpolation model;

[0118] Based on the basic Kriging interpolation model, sequence sampling optimization is performed to obtain enhanced sampling data;

[0119] A hierarchical proxy model is constructed based on enhanced sampling data, in which the global proxy model uses a second-order polynomial regression function as a trend term, and the local proxy model uses moving kriging to model the local area.

[0120] Based on the hierarchical agent model, the population is initialized, the wiring path parameters are encoded, and the time forgetting factor and space forgetting factor are set for each individual to obtain the encoded population;

[0121] The coding population is input into the genetic algorithm for crossover operation, and the Gaussian mutation operator is used for mutation operation to obtain the optimized population;

[0122] The optimized population is evaluated by multiple objectives, with stress minimization and path length minimization as the optimization objectives. The optimized population is stratified by fast non-dominated sorting, and the crowding distance of each individual is calculated to obtain a set of candidate solutions.

[0123] A fuzzy comprehensive evaluation is performed on the candidate solution set, a triangle membership function is established, and the optimal wiring path solution for the optical fiber distribution cabinet is selected according to the maximum membership principle.

[0124] Specifically, Latin hypercube sampling is performed on the wiring path prediction data to obtain initial training sample data. Latin hypercube sampling is a statistical method for sampling in high-dimensional parameter space, which is evenly distributed in the parameter space to ensure sampling coverage in each dimension. Latin hypercube sampling can avoid over-concentration or over-sparseness in high-dimensional space, thereby achieving higher sampling efficiency with fewer samples. In Latin hypercube sampling, the value range of each input variable is divided into several equal intervals, and then a sample point is randomly selected in each interval, and all input variables are combined. This method can effectively generate sample data containing all important variable information. Assuming that n input variables are sampled, Latin hypercube sampling can generate a sample matrix of the following form:

[0125] ;

[0126] in, It is Line The sample points of each column are evenly distributed in the corresponding input variable interval. is the sample size, is the number of input variables. In this way, Latin hypercube sampling provides uniformly distributed training samples for subsequent regression models. Input the initial training sample data into the Gaussian process regressor. Gaussian process regressor is a non-parametric regression method based on probabilistic inference, which is used to deal with data with uncertainty and noise. In Gaussian process regression, it is assumed that the objective function follows a Gaussian process in the input space, which is described by a mean function and a covariance function. Through the maximum likelihood estimation method, the optimal parameters in the model (such as the hyperparameters of the covariance function) are determined, and a basic Kriging interpolation model is established. The goal of the Gaussian process regressor is to infer the probability distribution of the output value through training data. The covariance function of the Gaussian process takes the following form:

[0127] ;

[0128] in, Is the input point and The covariance between is the signal variance, is the length scale, is the noise variance, is the Dirac delta function, which is used to represent noise. By maximizing the likelihood function, these hyperparameters are estimated, and these parameters are used to build a basic Kriging interpolation model. Based on the basic Kriging interpolation model, sequence sampling optimization is performed to obtain enhanced sampling data. Sequence sampling optimization is a sampling method that gradually increases samples based on the current sample data. By taking more samples in areas where the uncertainty of the current model prediction is large, the accuracy of the model is gradually improved to obtain more accurate interpolation results. Enhanced sampling data is a sample obtained by further sampling optimization on the basis of the original training data set, which can effectively improve the generalization ability of the model and avoid overfitting in model training. On this basis, a hierarchical surrogate model is constructed. The hierarchical surrogate model is a composite model composed of a global surrogate model and a local surrogate model. The global surrogate model uses a second-order polynomial regression function as a trend term to capture the main trend of the global data. Its model form is:

[0129] ;

[0130] in, is the regression coefficient to be estimated, is the input variable. Through this global model, the global relationship and trend between the input variables are captured. The local agent model uses moving kriging to model the local area. Moving kriging is a local weighted interpolation method that weights local samples within a certain range to establish a more refined model for a local area. This method can effectively handle the local nonlinear characteristics of the data and avoid the inaccurate fitting problem of the global model. Based on the hierarchical agent model, the population is initialized, the wiring path parameters are encoded, and the encoding representation of the individual is obtained. The encoding of each individual includes the starting coordinates, the end coordinates and the coordinates of the intermediate control points of the path. For each individual, the time forgetting factor and the space forgetting factor are set to dynamically adjust the contribution of each individual in the optimization process. The time forgetting factor and the space forgetting factor are used to represent the attenuation effect of the individual in the time dimension and the space dimension, respectively, so as to better adjust the search strategy of the genetic algorithm. With the introduction of these factors, the optimization process can be more flexible and accurate. The encoded population is input into the genetic algorithm for crossover operation, and the Gaussian mutation operator is used for mutation operation. Genetic algorithm is an optimization method that simulates natural selection and genetic mechanism, and gradually finds the optimal solution to the problem by simulating the evolution process of organisms. In the crossover operation, two individuals exchange some genes to produce new offspring individuals; in the mutation operation, the Gaussian mutation operator is used to perturb the genes of individuals to increase the diversity of the population. The mutation operation is performed through the following formula:

[0131] ;

[0132] in, It is a mutated individual. It is the original individual. represents a mean of 0 and a standard deviation of The normal distribution of is the step size of mutation. After the optimized population is obtained, a multi-objective evaluation is performed on it, with stress minimization and path length minimization as the optimization objectives. Multi-objective optimization problems involve the simultaneous optimization of multiple objectives, and it is necessary to comprehensively consider the trade-offs between different objectives. In this process, the optimized population is stratified by fast non-dominated sorting, and the crowding distance of each individual is calculated. After calculating the crowding distance of each individual, a set of candidate solutions is obtained. A fuzzy comprehensive evaluation is performed on the candidate solution set, a triangular membership function is established, and the optimal wiring path solution for the fiber optic distribution cabinet is selected by the maximum membership principle. The triangular membership function is used to describe uncertainty and ambiguity, and can judge the pros and cons of each path solution by defining different membership regions.

[0133] In a specific embodiment, the execution step is based on the hierarchical proxy model to initialize the population, encode the wiring path parameters, set the time forgetting factor and the space forgetting factor for each individual, and the process of obtaining the encoded population can specifically include the following steps:

[0134] The optimization variables of the hierarchical proxy model are encoded with real numbers, and the coordinates of the starting point, the end point and the intermediate control point of the wiring path are mapped to the interval [0,1] to obtain the normalized path parameters;

[0135] Generate population individuals based on normalized path parameters, use uniformly distributed random numbers to generate multiple initial individuals in the interval [0,1] to obtain the initial population;

[0136] The time dimension fitness value is calculated for each individual in the initial population, and the time forgetting factor is set based on the historical performance of the individual in the evolutionary process to obtain the time-weighted individual;

[0137] The spatial dimension fitness value is calculated for the time-weighted individuals, and the spatial forgetting factor is set based on the relative position of the individuals in the group to obtain the double forgetting factor individuals;

[0138] Sort the double forgetting factor individuals according to their fitness values ​​to obtain an elite population, and perform adaptive population size adjustment on the elite population to obtain a dynamic population;

[0139] A population distribution model is constructed based on a dynamic population. The individual distribution is modeled to obtain the population distribution characteristics. The individuals are clustered and grouped according to the population distribution characteristics, and local search is performed within each group to obtain the coded population.

[0140] Specifically, the optimization variables of the hierarchical proxy model are encoded with real numbers, and the starting coordinates, end coordinates, and intermediate control point coordinates of the wiring path are mapped to the interval [0,1] to obtain the normalized path parameters. Assume that the starting coordinates of the wiring path are The end point coordinates are , the coordinates of the middle control point are , these coordinates are normalized as follows:

[0141] ;

[0142] in, and are the minimum and maximum coordinate values, respectively, and the normalized parameter values Mapped to the [0,1] interval to ensure that the path parameters are optimized at the same scale. Generate population individuals based on the normalized path parameters, and use uniformly distributed random numbers to generate multiple initial individuals in the [0,1] interval to obtain an initial population, where each individual represents a possible wiring path parameter solution. In practical applications, uniformly distributed random numbers are used to generate the start, end, and control point coordinates of the path to ensure that the generated initial population covers the entire solution space. For example, if each path contains three control points, each individual is represented as a vector containing three coordinates:

[0143] ;

[0144] in, are all randomly generated coordinates in the interval [0,1]. For each individual in the initial population, calculate its fitness value in the time dimension. The fitness value measures the performance of an individual in a certain dimension, and the fitness value in the time dimension is calculated based on the individual's historical performance during the evolution process. To this end, a time forgetting factor is set to adjust the degree of dependence of the individual on historical performance during the optimization process. The introduction of the time forgetting factor makes the influence of the individual's historical fitness gradually decrease over time, so that the algorithm pays more attention to the performance of the current stage. For example, let the time forgetting factor be , whose value gradually decreases with the number of generations of individuals. Assume that the historical fitness of an individual is , then the individual's time-weighted fitness is expressed as:

[0145] ;

[0146] in, As the number of generations decreases, it means that the fitness of the more distant history has less impact. The fitness value of the spatial dimension is calculated for the time-weighted individual. The fitness value of the spatial dimension is set based on the relative position of the individual in the group. The spatial forgetting factor is used to adjust the degree of influence of the relative position of the individual in the group on the fitness. The role of the spatial forgetting factor is to make the algorithm more inclined to explore those individuals with large differences in the solution space during the search process. Assume that the spatial fitness of an individual is , and its spatial forgetting factor is , the double forgetting factor fitness of the individual (i.e., the fitness after time weighting and space weighting) is expressed as:

[0147] ;

[0148] By introducing the double forgetting factor, the optimization process can more effectively balance the information in both time and space, and avoid local search from falling into premature convergence. The population is sorted according to the fitness of the double forgetting factor individuals to obtain the elite population. The elite population is those individuals with higher fitness, which will be retained and participate in the next round of genetic operations. The selection of the elite population can ensure that excellent individuals will not be lost during the genetic process and accelerate the optimization process. The elite population is adaptively adjusted to meet the needs of the current optimization stage. By adjusting the size of the population, the diversity of the search is guaranteed to avoid over-convergence or insufficient samples. The size of the elite population is adjusted according to the fitness distribution of the current population. In this way, the size of the elite population is gradually adjusted as the number of iterations increases to meet the optimization needs of different stages. A population distribution model is constructed based on the dynamic population to model the individual distribution of the population. The population distribution model can provide a global view of the distribution characteristics of the population in the solution space, helping the genetic algorithm to achieve more effective exploration and development in the search process. The population distribution characteristics are described by calculating the density of each individual, that is, the ratio of the number of individuals in the area where the individual is located to the entire solution space. Based on the population distribution characteristics, the individuals are clustered and grouped. The clustering method can group similar individuals into one group and perform local search within each group, which speeds up the convergence of the algorithm and improves the accuracy of the search. When performing local search within each cluster group, the genetic algorithm will perform a fine-grained search to improve the fitness of individuals in each area and increase the coverage of the solution space. In this way, the search efficiency of the entire population is improved, while avoiding over-exploration of known areas. The resulting encoded population becomes the input for the next round of genetic operations. Through this series of operations, the population is gradually optimized in the search space and eventually finds the optimal path solution that meets the target requirements.

[0149] See also Figure 2 , Figure 2 A schematic block diagram of the structure of the optical fiber distribution cabinet automatic wiring path optimization device 200 provided in the embodiment of the present application, as shown in FIG. Figure 2 As shown, the optical fiber distribution cabinet automatic wiring path optimization device 200 includes:

[0150] Establishing module 210, for performing mesh division on the optical fiber distribution cabinet, establishing an optical fiber physical stress finite element model, and performing stress analysis and calculation on the optical fiber physical stress finite element model to obtain optical fiber node stress distribution data;

[0151] The analysis module 220 is used to perform correlation analysis between the node stress value and the path length based on the optical fiber node stress distribution data, and obtain the correlation data between the optical fiber stress and the wiring path;

[0152] An extraction module 230 is used to input the optical fiber stress and wiring path correlation data into an encoder-decoder structure for deep feature extraction to obtain a target feature parameter set;

[0153] A prediction module 240 is used to perform wiring path prediction on a target feature parameter set to obtain wiring path prediction data;

[0154] The solution module 250 is used to construct a Kriging proxy model based on the wiring path prediction data, and perform stress distribution optimization solution through a double forgetting factor genetic algorithm to obtain an optimal wiring path solution for the optical fiber wiring cabinet.

[0155] Through the collaborative cooperation of the above-mentioned components, by establishing a finite element model of optical fiber physical stress, combined with explicit dynamics solution and stress analysis, the stress distribution data of optical fiber nodes can be accurately obtained, effectively avoiding the stress damage problem caused by traditional empirical wiring; the encoder-decoder structure is used for deep feature extraction, and stress features and path features are integrated to achieve accurate modeling of optical fiber wiring paths, overcoming the problem of insufficient nonlinear feature extraction capabilities of traditional methods; the wiring path prediction method based on the dual support vector regression structure significantly improves the prediction accuracy through the synergy of the main regressor and the residual regressor, introduces the Kriging proxy model and the dual forgetting factor genetic algorithm, realizes the multi-objective optimization of the wiring path, and takes into account the optimization goals of minimizing stress and minimizing path length; the dynamic adjustment mechanism of the time forgetting factor and the space forgetting factor is used to enhance the global search capability of the algorithm and avoid local optimal solutions; the hierarchical proxy model and the adaptive population adjustment strategy are used to significantly improve the computational efficiency of the optimization algorithm, and it has strong robustness and generalization capabilities.

[0156] See also Figure 3 , Figure 3 The present application provides a schematic block diagram of the structure of an automated wiring path optimization device 300 for a fiber optic wiring cabinet, wherein the automated wiring path optimization device 300 for a fiber optic wiring cabinet includes a processor 301 and a memory 302, wherein the processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0157] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned methods for optimizing the automatic wiring path of the optical fiber wiring cabinet.

[0158] The processor 301 is used to provide computing and control capabilities to support the operation of the entire optical fiber distribution cabinet automatic wiring path optimization device 300.

[0159] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned methods for optimizing the automatic wiring path of the optical fiber wiring cabinet.

[0160] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the automatic wiring path optimization device 300 for the optical fiber distribution cabinet involved in the scheme of the present application. The specific automatic wiring path optimization device 300 for the optical fiber distribution cabinet may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0161] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0162] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the optical fiber distribution cabinet automatic wiring path optimization device 300 described above can refer to the corresponding process of the aforementioned optical fiber distribution cabinet automatic wiring path optimization method, and will not be repeated here.

[0163] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the method for optimizing the automatic wiring path of the optical fiber wiring cabinet provided in the embodiment of the present application.

[0164] The computer-readable storage medium may be an internal storage unit of the optical fiber distribution cabinet automatic wiring path optimization device 300 of the aforementioned embodiment, such as a hard disk or memory of the optical fiber distribution cabinet automatic wiring path optimization device 300. The computer-readable storage medium may also be an external storage device of the optical fiber distribution cabinet automatic wiring path optimization device 300, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the optical fiber distribution cabinet automatic wiring path optimization device 300.

[0165] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0166] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0167] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for optimizing the automatic wiring path of an optical fiber wiring cabinet, characterized in that: include: Meshing the optical fiber distribution cabinet, establishing an optical fiber physical stress finite element model, and performing stress analysis and calculation on the optical fiber physical stress finite element model to obtain optical fiber node stress distribution data; Based on the optical fiber node stress distribution data, a correlation analysis is performed between the node stress value and the path length to obtain optical fiber stress and wiring path correlation data; Inputting the optical fiber stress and wiring path correlation data into an encoder-decoder structure for deep feature extraction to obtain a target feature parameter set; Performing wiring path prediction on the target characteristic parameter set to obtain wiring path prediction data; Latin hypercube sampling is performed on the wiring path prediction data to obtain initial training sample data; the initial training sample data is input into a Gaussian process regressor, and the optimal parameter value is determined by the maximum likelihood estimation method to obtain a basic Kriging interpolation model; sequence sampling optimization is performed based on the basic Kriging interpolation model to obtain enhanced sampling data; a hierarchical proxy model is constructed based on the enhanced sampling data, wherein the global proxy model uses a second-order polynomial regression function as a trend term, and the local proxy model uses the moving Kriging method to model the local area; based on the hierarchical proxy model, a population is initialized, the wiring path parameters are encoded, and a time forgetting factor and a space forgetting factor are set for each individual to obtain a coded population; the coded population is input into a genetic algorithm for a crossover operation, and a Gaussian mutation operator is used to perform a mutation operation to obtain an optimized population; A multi-objective evaluation is performed on the optimized population, with stress minimization and path length minimization as the optimization objectives. The optimized population is stratified using fast non-dominated sorting, and the crowding distance of each individual is calculated to obtain a set of candidate solutions. A fuzzy comprehensive evaluation is performed on the set of candidate solutions, a triangle membership function is established, and the optimal wiring path solution of the optical fiber distribution cabinet is selected according to the maximum membership principle.

2. The method for optimizing the automatic wiring path of an optical fiber distribution cabinet according to claim 1, characterized in that: The gridding of the optical fiber distribution cabinet is performed, a finite element model of optical fiber physical stress is established, and stress analysis and calculation are performed on the finite element model of optical fiber physical stress to obtain optical fiber node stress distribution data, including: Performing three-dimensional modeling and hexahedral unit division on the optical fiber distribution cabinet to obtain grid unit data, and adding optical fiber material property parameters to the grid unit data to obtain an initialized finite element model; Applying gravity load and fixed constraints to the initialized finite element model to obtain constraint boundary conditions, and constructing an optical fiber physical stress finite element model based on the constraint boundary conditions to obtain a finite element model to be analyzed; Importing the finite element model to be analyzed into an explicit dynamics solver, performing node stress calculation to obtain initial node stress data, and performing optical fiber bending stress analysis on the initial node stress data to obtain bending stress data; The optical fiber vibration characteristic analysis is performed based on the bending stress data to obtain vibration characteristic data, and the initial node stress data is corrected according to the vibration characteristic data to obtain optical fiber node stress distribution data.

3. The method for optimizing the automatic wiring path of an optical fiber distribution cabinet according to claim 2, characterized in that: The correlation analysis between the node stress value and the path length is performed based on the optical fiber node stress distribution data to obtain the optical fiber stress and wiring path correlation data, including: Performing Fourier transformation on the stress distribution data of the optical fiber node to obtain stress spectrum data, and extracting the main frequency component of the stress spectrum data to obtain the main frequency stress characteristics; Performing static wiring on the optical fiber distribution cabinet, collecting node stress values ​​and distribution path length data to obtain a static data set, and dynamically adjusting the optical fiber distribution cabinet, collecting real-time stress and path length change data to obtain a dynamic data set; The static data set and the dynamic data set are combined to construct a polynomial fitting sample to obtain fitting training data, and the fitting training data is normalized to obtain standardized training data; The polynomial coefficients are calculated based on the standardized training data using the least square method to obtain a polynomial fitting model, and the main frequency stress characteristics are input into the polynomial fitting model to perform stress path correlation calculation to obtain optical fiber stress and wiring path correlation data.

4. The method for optimizing the automatic wiring path of an optical fiber distribution cabinet according to claim 3, characterized in that: The optical fiber stress and wiring path association data are input into the encoder-decoder structure for deep feature extraction to obtain a target feature parameter set, including: Segmenting the optical fiber stress and wiring path correlation data to obtain stress characteristic data and path characteristic data; Inputting the stress feature data into the first encoder in the encoder-decoder structure for feature compression, the first encoder comprises 5 convolutional layers, each convolutional layer uses a convolution kernel with a step size of 2 and a LeakyReLU activation function, and each convolutional layer is followed by a batch normalization layer to obtain stress compression features; Inputting the path feature data into a second encoder in the encoder-decoder structure for feature compression, wherein the encoder structure and parameters of the second encoder are the same as those of the first encoder, to obtain a path compression feature; The stress compression feature and the path compression feature are feature-fused to obtain fused feature data, and the fused feature data is input into a decoder in the encoder-decoder structure for feature reconstruction, wherein the decoder comprises 5 deconvolution layers, each deconvolution layer uses a deconvolution kernel with a step size of 2 and a ReLU activation function, and each deconvolution layer is followed by a batch normalization layer and a residual connection to obtain reconstructed feature data; The reconstructed feature data is subjected to spatial attention enhancement to obtain enhanced feature data, and multi-scale feature extraction is performed on the enhanced feature data to obtain a target feature parameter set.

5. The method for optimizing the automatic wiring path of an optical fiber wiring cabinet according to claim 4, characterized in that: The performing wiring path prediction on the target feature parameter set to obtain wiring path prediction data includes: Inputting the target feature parameter set into a radial basis kernel function for nonlinear mapping to obtain kernel feature data; Based on the kernel feature data, a penalty factor and a precision parameter of a support vector regression machine are determined by a grid search method to obtain optimized hyperparameters; Constructing a dual support vector regression structure based on the optimized hyperparameters, wherein the dual support vector regression structure includes a main regressor and a residual regressor; The kernel feature data is input into the main regressor for primary regression calculation, and the main regressor adopts an ε-insensitive loss function to obtain a main prediction result; Performing residual calculation on the main prediction result and the kernel feature data to obtain residual feature data; The residual feature data is input into the residual regressor for secondary regression calculation, and the residual regressor adopts Huber loss function to obtain the residual prediction result; The main prediction result and the residual prediction result are weightedly fused to obtain wiring path prediction data.

6. The method for optimizing the automatic wiring path of an optical fiber distribution cabinet according to claim 1, characterized in that: The population is initialized based on the hierarchical proxy model, wiring path parameters are encoded, and a time forgetting factor and a space forgetting factor are set for each individual to obtain an encoded population, including: The optimization variables of the hierarchical proxy model are encoded with real numbers, and the starting point coordinates, the end point coordinates and the intermediate control point coordinates of the wiring path are mapped to the interval [0,1] to obtain normalized path parameters; Generate population individuals based on the normalized path parameters, and use uniformly distributed random numbers to generate multiple initial individuals in the interval [0,1] to obtain an initial population; Calculating the time dimension fitness value for each individual of the initial population, setting a time forgetting factor based on the historical performance of the individual in the evolution process, and obtaining a time-weighted individual; Calculating the fitness value of the spatial dimension for the time-weighted individual, setting a spatial forgetting factor based on the relative position of the individual in the group, and obtaining a double forgetting factor individual; The double forgetting factor individuals are sorted according to fitness values ​​to obtain an elite population, and the elite population is adaptively adjusted to obtain a dynamic population; A population distribution model is constructed based on the dynamic population, individual distribution is modeled to obtain population distribution characteristics, individuals are clustered and grouped according to the population distribution characteristics, and local search is performed within each group to obtain a coded population.

7. An automatic wiring path optimization device for an optical fiber wiring cabinet, characterized in that: The method for optimizing the automatic wiring path of an optical fiber wiring cabinet according to any one of claims 1 to 6 comprises: Establishing a module for meshing the optical fiber distribution cabinet, establishing an optical fiber physical stress finite element model, and performing stress analysis and calculation on the optical fiber physical stress finite element model to obtain optical fiber node stress distribution data; An analysis module, used to perform correlation analysis between node stress value and path length based on the optical fiber node stress distribution data, and obtain correlation data between optical fiber stress and wiring path; An extraction module, used for inputting the optical fiber stress and wiring path correlation data into an encoder-decoder structure for deep feature extraction to obtain a target feature parameter set; A prediction module, used to perform wiring path prediction on the target characteristic parameter set to obtain wiring path prediction data; A solution module is used to perform Latin hypercube sampling on the distribution path prediction data to obtain initial training sample data; input the initial training sample data into a Gaussian process regressor, determine the optimal parameter value by maximum likelihood estimation, and obtain a basic Kriging interpolation model; perform sequence sampling optimization based on the basic Kriging interpolation model to obtain enhanced sampling data; construct a hierarchical proxy model based on the enhanced sampling data, wherein the global proxy model uses a second-order polynomial regression function as a trend term, and the local proxy model uses a moving Kriging method to model the local area; perform population initialization based on the hierarchical proxy model, and perform optimization on the distribution path. The parameters are encoded, and a time forgetting factor and a space forgetting factor are set for each individual to obtain a coded population; the coded population is input into a genetic algorithm for a crossover operation, and a Gaussian mutation operator is used for a mutation operation to obtain an optimized population; a multi-objective evaluation is performed on the optimized population, and stress minimization and path length minimization are used as optimization objectives. The optimized population is stratified by using a fast non-dominated sorting, and the crowding distance of each individual is calculated to obtain a set of candidate solutions; a fuzzy comprehensive evaluation is performed on the set of candidate solutions, a triangle membership function is established, and the optimal wiring path solution of the optical fiber distribution cabinet is selected by the maximum membership principle.

8. An automatic wiring path optimization device for optical fiber wiring cabinet, characterized in that: The optical fiber distribution cabinet automatic wiring path optimization device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the optical fiber distribution cabinet automatic wiring path optimization device to execute the optical fiber distribution cabinet automatic wiring path optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for optimizing the automatic wiring path of the optical fiber wiring cabinet according to any one of claims 1 to 6 is implemented.

Images