Optical fiber coupling efficiency improving method based on bat algorithm

Abstract

The invention discloses an optical fiber coupling efficiency improving method based on a bat algorithm. The method comprises the following steps that optical power data are collected and preprocessed; constructing a coupling parameter search space model; initializing an improved submerged space enhancement-bat algorithm, and outputting an initial coupling parameter vector; introducing countercurrent coding to generate a submerged space vector; calculating and outputting an optimal action vector through a differentiable strategy flow mapping mechanism; obtaining a current iteration optimal bat individual; outputting a coupling parameter vector of the global optimal bat individual; and the improved submerged space enhancement-bat algorithm is recalled to perform parameter optimization calculation, so that the optimization speed and the global optimal stability of optical fiber coupling are effectively improved, the coupling loss is remarkably reduced, and the automatic adjustment precision and the optical power stability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical communication control, and particularly relates to a method for improving fiber coupling efficiency based on a bat algorithm. BACKGROUND

[0002] As a key component in optical communication, optical sensing and optoelectronic test equipment, the coupling efficiency of the fiber coupling system directly affects the optical power transmission performance and signal transmission stability of the system. In the application scenarios of fiber devices, fiber arrays and photonic chip packaging, how to realize high-precision and high-robustness automatic fiber coupling has always been an important research direction in the field of optical manufacturing and control. Traditional fiber coupling usually adopts a combination of mechanical adjustment and empirical parameter adjustment, and is manually optimized through a displacement platform and an optical power detection device. Although this method is simple to implement, it is time-consuming and has poor repeatability, and is extremely sensitive to environmental disturbances, which makes it difficult to meet the requirements of high precision and real-time performance.

[0003] With the development of intelligent control and optimization algorithms, researchers have gradually introduced intelligent optimization algorithms into the fiber coupling process to realize global optimization search of multiple parameters. At present, commonly used intelligent optimization algorithms include particle swarm optimization algorithm, genetic algorithm, ant colony algorithm, etc. These algorithms can improve the automation level and search efficiency of fiber coupling to some extent. However, the existing algorithms have the following shortcomings: first, the traditional bat algorithm is prone to local optimum in high-dimensional coupling parameter space, especially in a multi-parameter nonlinear coupling system, the algorithm has slow convergence speed and insufficient global optimization ability; second, the optical power signal in the fiber coupling system is often accompanied by noise fluctuations and dynamic drift, resulting in large fluctuations in the fitness function, and the traditional optimization algorithm is prone to oscillation and unstable search during the iteration update process; third, the existing algorithms do not fully consider the linear constraints and physical feasibility between the coupling parameters, and the generated parameter solution may exceed the actual working range of the system, affecting the practicability of the algorithm. SUMMARY

[0004] One object of the present application is to provide a method for improving fiber coupling efficiency based on a bat algorithm. The present application makes full use of optical power signal preprocessing technology, coupling parameter search space modeling technology, latent space reversible flow coding mechanism and reinforcement strategy network structure, and introduces an improved latent space reinforcement-bat algorithm in the fiber coupling process to dynamically optimize the fiber coupling parameters and realize intelligent coupling control with high efficiency, low error and strong robustness.

[0005] According to the method for improving fiber coupling efficiency based on a bat algorithm, the method comprises the following steps: Collecting and preprocessing the optical power data of the fiber coupling system; According to the pre-processed optical power data, the coupling parameter vector and the value range of the fiber coupling system are set, and a coupling parameter search space model is constructed; Based on the coupling parameter search space model, an improved latent space reinforcement-bat algorithm is initialized, and an initial coupling parameter vector is output; The initial coupling parameter vector is input into the latent space coding module of the improved latent space reinforcement-bat algorithm, and a latent space vector is generated by introducing reversible flow coding; The latent space vector is input into the latent space reinforcement strategy network module of the improved latent space reinforcement-bat algorithm, and the optimal action vector is calculated by the differentiable policy flow mapping mechanism; The optimal action vector is input into the bat individual evolution update module of the improved latent space reinforcement-bat algorithm, and the current iteration optimal bat individual is obtained; The current iteration optimal bat individual is input into the adaptive adjustment and termination control module of the improved latent space reinforcement-bat algorithm, the loudness and pulse emission rate of the current iteration optimal bat individual are adaptively corrected, and the coupling parameter vector of the global optimal bat individual is output by adjusting the latent space reinforcement strategy parameters according to the feedback signal of the optical power data; According to the coupling parameter vector of the global optimal bat individual, the optical fiber coupling system is adjusted to the corresponding coupling parameter state and locked by controlling the actuator, and the optical power signal is continuously monitored during the operation of the optical fiber coupling system. When the detected optical power signal is less than the preset threshold, the improved latent space reinforcement-bat algorithm is called again for parameter optimization calculation.

[0006] Optionally, the preprocessing includes noise filtering, smoothing processing and normalization preprocessing.

[0007] Optionally, the construction of the coupling parameter search space model specifically includes: Based on the pre-processed optical power signal, the optical power response features are extracted, including the optical power amplitude change rate, the coupling loss change trend and the system steady-state power distribution; The coupling parameter vector is defined, and the lower limit and upper limit of each parameter component are set according to the change interval of the optical power response features to form a lower limit vector and an upper limit vector. The coupling parameter vector includes the optical fiber end position parameter, the optical fiber end angle parameter, and the optical component spacing parameter; The coupling parameter feasible region is established based on the lower limit vector and the upper limit vector, and the linear dependence relationship between each coupling parameter component and the system feasible state are limited by the linear constraint matrix and the linear constraint vector to obtain the feasible region boundary model; Under the constraint condition of the feasible region boundary model, the parameter resolution step is set according to the dynamic variation range of the preprocessed optical power data and the coupling parameter sensitivity, and the boundary information of the lower value vector and the upper value vector is used for step quantization and normalization processing of each component of the coupling parameter vector, so as to obtain a quantized coupling parameter vector and a normalized coupling parameter vector; The lower value vector, the upper value vector, the linear constraint matrix, the linear constraint vector, the parameter resolution step, the quantized coupling parameter vector and the normalized coupling parameter vector are combined to form a coupling parameter search space model.

[0008] Optionally, the output of the initial coupling parameter vector specifically includes: Based on the coupling parameter search space model, an improved latent space reinforcement-bat algorithm is loaded and initialized, and initialization parameters are set, the initialization parameters including a population size, a maximum iteration number, a search frequency range, an initial loudness and an initial pulse emission rate; Based on the initialization parameters, projection is performed under the constraint of the linear constraint matrix and the linear constraint vector of the coupling parameter search space model, to generate a candidate coupling parameter vector, the generation of the candidate coupling parameter vector being that a coupling parameter sample is generated according to the initialization parameters, the coupling parameter sample is substituted into the constraint condition defined by the linear constraint matrix and the linear constraint vector, and the coupling parameter sample that does not satisfy the constraint condition is modified to obtain a coupling parameter sample that satisfies all the constraint conditions as the candidate coupling parameter vector; For the candidate coupling parameter vector, an initial coupling parameter vector is generated for each bat individual within the limited range of the lower value vector and the upper value vector.

[0009] Optionally, the generation of the latent space vector specifically includes: The initial coupling parameter vector is normalized to map each parameter value of the initial coupling parameter vector to a normalized interval of zero to one in proportion, to generate a normalized coupling parameter vector; Based on the normalized coupling parameter vector, a coupling parameter correlation matrix is constructed, and a coupling parameter sensitivity coefficient is calculated, the construction of the coupling parameter correlation matrix being that covariance analysis is performed on the normalized values of each coupling parameter under multiple groups of optical power data, linear correlation coefficients and nonlinear coupling degrees between parameter components are calculated, and the coupling parameter correlation matrix is formed by row and column combination, the calculation process of the coupling parameter sensitivity coefficient being that a small disturbance is sequentially applied to each component of the normalized coupling parameter vector, the optical power output change before and after the disturbance is recorded, the ratio of the optical power output change amount to the corresponding parameter disturbance amount is compared, the local response degree of the component to the optical power output is obtained, and the ratio is defined as the coupling parameter sensitivity value of the component; The normalized coupling parameter vector is input into a latent space coding module of the improved latent space reinforcement-bat algorithm, a weighted normalized coupling parameter vector is obtained by performing a weighted transformation on the normalized coupling parameter vector through a feature weighting layer; The weighted normalized coupling parameter vector is subjected to nonlinear feature mapping through a reversible flow coding to output a latent space vector, which refers to a state of a bat individual in a latent space feature domain.

[0010] Optionally, the output of the optimal action vector specifically includes: The latent space vector is input into a latent space reinforcement strategy network module of the improved latent space reinforcement-bat algorithm, a latent space feature is obtained by performing linear mapping and nonlinear activation calculation through a differentiable policy flow mapping mechanism, the latent space reinforcement strategy network module includes an input layer, a hidden layer and an action generation layer; The latent space feature is input into the hidden layer of the latent space reinforcement strategy network module to perform feature mapping through a weight matrix and apply a nonlinear activation function, and an original action vector is output; An amplitude constraint processing is performed on the original action vector to obtain a standardized action output; The standardized action output is divided into a frequency adjustment amount, a speed correction amount, a position disturbance amount, a loudness adjustment amount and a pulse emission rate adjustment amount according to the parameter categories, and the components are subjected to amplitude correction and constraint according to a preset scaling coefficient and a mapping rule to obtain an optimal action vector.

[0011] Optionally, the obtaining of the current iteration optimal bat individual specifically includes: The search frequency of the bat individual is updated according to the frequency adjustment amount in the optimal action vector to obtain an updated frequency adjustment amount, the updated frequency adjustment amount is obtained by superimposing the search frequency and the frequency adjustment amount of the last iteration; The speed vector of the bat individual is updated according to the speed correction amount in the optimal action vector and the coupling parameter vector of the optimal bat individual to obtain an updated speed vector, the updated speed vector is obtained by summing the original speed vector and the speed correction amount, and the coupling parameter difference between the bat individual and the global optimal individual is weighted and corrected by combining the search frequency; The coupling parameter vector of the bat individual is updated according to the updated speed vector and the position disturbance amount in the optimal action vector to obtain an updated position vector, the updated position vector is obtained by the original position vector, the updated speed vector and the position disturbance amount; The updated frequency adjustment amount, the updated speed vector and the updated position vector are spliced to form an updated coupling parameter vector of the bat individual; The updated bat individual coupling parameter vector is input into the optical fiber coupling system, the corresponding optical power output signal is collected by controlling the actuator, the fitness value of the bat individual is calculated, the bat individual with the highest fitness value is selected as the optimal bat individual of the current iteration, and the calculation process of the fitness value of the bat individual is based on the intensity of the optical power output signal. The collected optical power signal is compared with the target maximum optical power, the minimum coupling loss is taken as the evaluation criterion, and the fitness value of each bat individual is calculated.

[0012] Optionally, the output of the coupling parameter vector of the global optimal bat individual specifically includes: The loudness adjustment amount and the pulse emission rate adjustment amount of the optimal bat individual of the current iteration are adaptively corrected to obtain a correction result, and the correction result is updated to the parameter state of the improved latent space reinforcement-bat algorithm; The strategy parameters of the latent space reinforcement strategy network module are adaptively updated through the updated parameter state of the improved latent space reinforcement-bat algorithm, to form updated strategy parameters, and the updating process is performed by adjusting the strategy parameters in the direction of increasing reward value. According to the updated strategy parameters, a termination condition is determined, and the maximum number of iterations is set as a preset upper limit. When the number of iterations reaches the preset upper limit, the coupling parameter vector of the global optimal bat individual is output.

[0013] Optionally, the re-calling of the improved latent space reinforcement-bat algorithm specifically includes: According to the coupling parameter vector of the global optimal bat individual, the coupling parameter state of the optical fiber coupling system is adjusted to the state corresponding to the global optimal coupling parameter vector by controlling the actuator, and a coupling parameter state locking operation is performed to form a global optimal coupling state. Based on the global optimal coupling state, an online monitoring parameter of the optical power signal is set, the optical power signal is collected according to the set sampling period, and the sliding average of the optical power is calculated in each monitoring window. The online monitoring parameters include the sampling period, the monitoring window length, the power threshold and the power drift threshold. When the sliding average of the optical power is less than the preset power threshold, a re-optimization trigger signal is generated. The generation of the re-optimization trigger signal is based on the optical power real-time monitoring module. When the sliding average optical power is lower than the preset power threshold in the continuous sampling period, the monitoring module determines that the system coupling state deviates from the optimal interval, and automatically outputs a digital trigger signal to the control unit. When the re-optimization trigger signal is detected to be in the active state, the coupling parameter state of the current optical fiber coupling system is read and input to the control unit together with the current optical power signal. An optimization start instruction is generated in the control unit, the optimization process of the improved latent space reinforcement-bat algorithm is re-called, and parameter re-optimization calculation is performed.

[0014] The beneficial effects of this invention are: This invention achieves significant technical results by introducing an improved latent space reinforcement-bat algorithm into the fiber optic coupling process, realizing intelligent global optimization and dynamic adaptive adjustment of coupling parameters. The method uses optical power signals as optimization feedback. First, it establishes a search space model of the coupling parameters of the fiber optic coupling system. Then, it utilizes a reversible flow coding mechanism to map high-dimensional parameters into the latent space for compressed expression, effectively reducing coupling interference and search complexity. The algorithm executes reinforcement learning-style policy flow mapping within the latent space. Through the optimal action output of the policy network, it achieves efficient evolutionary updates of bat individuals in parameters such as frequency, speed, and position, making the search process more global and directional.

[0015] By dynamically feeding back optical power signals and adaptively correcting strategy parameters, this invention can adjust action outputs in real time during iteration and automatically correct coupling errors, thereby achieving rapid and stable convergence in multi-parameter nonlinear systems involving fiber end position, angle, and optical component spacing. This method ensures the physical feasibility of parameter values ​​during initialization, implements reinforcement learning-driven optimal actions during evolution, and achieves systematic control of globally optimal parameters during output, effectively avoiding the problems of traditional bat algorithms such as getting trapped in local optima, search oscillations, and insufficient convergence accuracy.

[0016] Compared to existing fiber coupling optimization methods based on genetic algorithms, particle swarm optimization, and traditional bat algorithms, this invention introduces latent space feature representation and differentiable strategy mapping mechanisms into the algorithm structure, enabling the search process to have self-learning and self-evolution capabilities, significantly improving the response speed and optimization accuracy of optical power signals. By maintaining the consistency of the mapping between latent and original spaces through reversible flow coding, the continuity and reversibility of parameter updates are strengthened, ensuring the stability and physical realizability of the optimization results.

[0017] In summary, this invention achieves intelligent optimization of the entire process of fiber coupling, from parameter modeling, feature mapping, strategy output to adaptive adjustment, significantly improving fiber coupling efficiency and system robustness. It has advantages such as fast convergence speed, strong global optimization capability, excellent noise resistance and high adaptability, providing an efficient, stable and learning-capable solution for automatic coupling control of complex fiber systems. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0019] Fig. 1A flow chart of a fiber coupling efficiency improvement method based on a bat algorithm is provided for the present application. Fig. 2 A module structure schematic diagram of an improved latent space reinforcement-bat algorithm of a fiber coupling efficiency improvement method based on a bat algorithm is provided for the present application. DETAILED DESCRIPTION

[0020] The present application will now be further described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams and only schematically illustrate the basic structure of the present application, and thus only show the components related to the present application.

[0021] REFERENCE Figs. 1-2 A fiber coupling efficiency improvement method based on a bat algorithm, comprising the following steps: Collecting optical power data of a fiber coupling system and preprocessing; According to the preprocessed optical power data, setting the coupling parameter vector and value range of the fiber coupling system, and constructing a coupling parameter search space model; Based on the coupling parameter search space model, initializing the improved latent space reinforcement-bat algorithm, and outputting the initial coupling parameter vector; Input the initial coupling parameter vector into the latent space encoding module of the improved latent space reinforcement-bat algorithm, and introduce reversible flow encoding to generate a latent space vector; Input the latent space vector into the latent space reinforcement strategy network module of the improved latent space reinforcement-bat algorithm, and calculate the output optimal action vector through the differentiable policy flow mapping mechanism; Input the optimal action vector into the bat individual evolution update module of the improved latent space reinforcement-bat algorithm, and obtain the current iteration optimal bat individual; Input the current iteration optimal bat individual into the adaptive adjustment and termination control module of the improved latent space reinforcement-bat algorithm, and adaptively correct the loudness and pulse emission rate of the current iteration optimal bat individual, and adjust the latent space reinforcement strategy parameters according to the feedback signal of the optical power data, and output the coupling parameter vector of the global optimal bat individual; According to the coupling parameter vector of the global optimal bat individual, control the actuator to adjust the fiber coupling system to the corresponding coupling parameter state and lock, and continuously monitor the optical power signal during the operation of the fiber coupling system, and when the detected optical power signal is less than the preset threshold, re-call the improved latent space reinforcement-bat algorithm for parameter optimization calculation.

[0022] In the present embodiment, the preprocessing includes noise filtering, smoothing processing and normalization preprocessing.

[0023] In the present embodiment, the construction of the coupling parameter search space model specifically includes: Based on the pre-processed optical power signal, an optical power response feature is extracted, the optical power response feature including an optical power amplitude change rate, a coupling loss change trend, and a system steady-state power distribution; A coupling parameter vector is defined, and a lower limit and an upper limit of each parameter component are set according to a variation interval of the optical power response feature, forming a lower limit value vector and an upper limit value vector, the coupling parameter vector including a fiber end position parameter, a fiber end angle parameter, and an optical component spacing parameter; A coupling parameter feasible region is established based on the lower limit value vector and the upper limit value vector, and a linear dependence relationship between each coupling parameter component and a system feasible state are defined by a linear constraint matrix and a linear constraint vector, to obtain a feasible region boundary model; Under the constraint condition of the feasible region boundary model, a parameter resolution step is set according to a dynamic variation range of the pre-processed optical power data and a coupling parameter sensitivity, and a step quantization and normalization processing is performed on each component of the coupling parameter vector by using the boundary information of the lower limit value vector and the upper limit value vector, to obtain a quantized coupling parameter vector and a normalized coupling parameter vector; The lower limit value vector, the upper limit value vector, the linear constraint matrix, the linear constraint vector, the parameter resolution step, the quantized coupling parameter vector, and the normalized coupling parameter vector are combined to form a coupling parameter search space model.

[0024] In the embodiment, the output of the initial coupling parameter vector specifically includes: Based on the coupling parameter search space model, an improved latent space reinforcement-bat algorithm is loaded and initialized, and initialization parameters are set, the initialization parameters including a population size, a maximum number of iterations, a search frequency range, an initial loudness, and an initial pulse emission rate; Based on the initialization parameters, a projection is performed under the constraint of the linear constraint matrix and the linear constraint vector of the coupling parameter search space model, to generate a candidate coupling parameter vector, the generation of the candidate coupling parameter vector being that a coupling parameter sample is generated according to the initialization parameters, the coupling parameter sample is substituted into the constraint condition defined by the linear constraint matrix and the linear constraint vector, coupling parameter samples that do not satisfy the constraint condition are modified, and a coupling parameter sample that satisfies all the constraint conditions is obtained as the candidate coupling parameter vector; For the candidate coupling parameter vector, an initial coupling parameter vector is generated for each bat individual within the limited range of the lower limit value vector and the upper limit value vector.

[0025] In the embodiment, the generation of the latent space vector specifically includes: The initial coupling parameter vector is normalized, each parameter value of the initial coupling parameter vector is proportionally mapped to a normalized interval of zero to one, and a normalized coupling parameter vector is generated; The coupling parameter correlation matrix is constructed based on the normalized coupling parameter vector, and the coupling parameter sensitivity coefficient is calculated. The construction of the coupling parameter correlation matrix is to perform covariance analysis on the normalized values of each coupling parameter under multiple groups of optical power data, calculate the linear correlation coefficient and the nonlinear coupling degree between each parameter component, and form the coupling parameter correlation matrix according to row and column combination. The calculation process of the coupling parameter sensitivity coefficient is to sequentially apply a small disturbance to each component of the normalized coupling parameter vector, record the changes of the optical power output before and after the disturbance, and obtain the local response degree of the component to the optical power output by comparing the ratio of the optical power output change amount to the corresponding parameter disturbance amount. The ratio is defined as the coupling parameter sensitivity value of the component. The normalized coupling parameter vector and the coupling parameter sensitivity coefficient are input into the latent space coding module of the improved latent space reinforcement-bat algorithm, and the normalized coupling parameter vector is weighted and transformed through the feature weighting layer to obtain a weighted normalized coupling parameter vector. The weighted normalized coupling parameter vector is nonlinearly mapped using reversible flow coding to output a latent space vector, which refers to the state of a bat individual in the latent space feature domain.

[0026] In this embodiment, the output of the optimal action vector specifically includes: The latent space vector is input into the latent space reinforcement strategy network module of the improved latent space reinforcement-bat algorithm, and linear mapping and nonlinear activation calculation are performed through the differentiable policy flow mapping mechanism to obtain latent space features. The latent space reinforcement strategy network module includes an input layer, a hidden layer, and an action generation layer. The latent space features are input into the hidden layer of the latent space reinforcement strategy network module for feature mapping through a weight matrix and applying a nonlinear activation function, and an original action vector is output. The original action vector is subjected to amplitude constraint processing to obtain a standardized action output. The standardized action output is divided into frequency adjustment, speed correction, position disturbance, loudness adjustment, and pulse emission rate adjustment according to the parameter categories, and the components are subjected to amplitude correction and constraint according to the preset scaling coefficients and mapping rules to obtain an optimal action vector.

[0027] In this embodiment, the current iteration optimal bat individual is obtained specifically by: The search frequency of the bat individual is updated according to the frequency adjustment in the optimal action vector to obtain an updated frequency adjustment, which is obtained by superimposing the search frequency and the frequency adjustment of the last iteration. According to the coupling parameter difference between the bat individual and the global optimal individual, the coupling parameter vector of the bat individual is updated by weighting correction combined with the search frequency to form a coupling parameter vector of an updated bat individual; According to the updated speed vector and the position disturbance in the optimal action vector, the coupling parameter vector of the bat individual is updated to obtain an updated position vector, and the updated position vector is obtained from the original position vector, the updated speed vector and the position disturbance; The updated frequency adjustment, the updated speed vector and the updated position vector are spliced to form an updated coupling parameter vector of the bat individual; The updated coupling parameter vector of the bat individual is input into the fiber coupling system, the corresponding optical power output signal is collected by controlling the actuator, and the fitness value of the bat individual is calculated, and the bat individual with the highest fitness value is selected as the optimal bat individual of the current iteration, and the calculation process of the fitness value of the bat individual is based on the intensity of the optical power output signal, the collected optical power signal is compared with the target maximum optical power, and the minimum coupling loss is taken as the evaluation criterion to calculate the fitness value of each bat individual.

[0028] In the embodiment, the output of the coupling parameter vector of the global optimal bat individual specifically includes: The loudness adjustment and the pulse emission rate adjustment of the current iteration optimal bat individual are adaptively corrected to obtain a correction result, and the correction result is updated to the parameter state of the improved latent space reinforcement-bat algorithm; The strategy parameters of the latent space reinforcement strategy network module are adaptively updated through the updated parameter state of the improved latent space reinforcement-bat algorithm to form updated strategy parameters, and the updating process is performed by adjusting the strategy parameters in the direction of increasing reward value; According to the updated strategy parameters, a termination condition is determined, and the maximum iteration number is set as a preset upper limit. When the iteration number reaches the preset upper limit, the coupling parameter vector of the global optimal bat individual is output.

[0029] In the embodiment, the re-calling of the improved latent space reinforcement-bat algorithm specifically includes: According to the coupling parameter vector of the global optimal bat individual, the coupling parameter state of the fiber coupling system is adjusted to the state corresponding to the global optimal coupling parameter vector by controlling the actuator, and a coupling parameter state locking operation is performed to form a global optimal coupling state. Based on the global optimal coupling state, the online monitoring parameters of the optical power signal are set, the optical power signal is collected according to the set sampling period, the sliding average value of the optical power is calculated in each monitoring window, and the online monitoring parameters include the sampling period, the monitoring window length, the power threshold and the power drift threshold; When the sliding average value of the optical power is less than the preset power threshold, a re-optimization trigger signal is generated, and the generation of the re-optimization trigger signal is based on the optical power real-time monitoring module. When the sliding average optical power is lower than the preset power threshold in a continuous sampling period, the monitoring module determines that the system coupling state deviates from the optimal interval, and automatically outputs a digital trigger signal to the control unit. When it is detected that the re-optimization trigger signal is in an activated state, the coupling parameter state of the current optical fiber coupling system is read and input to the control unit together with the current optical power signal. An optimization start instruction is generated in the control unit, the optimization process of the improved latent space reinforcement-bat algorithm is called again, and parameter re-optimization calculation is performed.

[0030] Embodiment 1 An optical fiber automatic coupling experimental system is taken as an implementation object. The experimental system is composed of a light source module, a collimating lens assembly, a six-dimensional precision displacement platform, a fiber clamping device, an optical power detection unit and a control computer. The light source is a semiconductor laser with a wavelength of 1550 nm, and the output power is stabilized at 10 mW. The optical fiber is a single-mode optical fiber SMF-28e, and the optical fiber end face has been polished by FC / APC end face polishing. The detection end uses a power meter Agilent 8163B with a measurement accuracy of ±0.01 dB.

[0031] In the experimental process, first, the system optical power data is collected and subjected to mean filtering and normalization preprocessing to remove high-frequency noise. Then, a coupling parameter search space model is constructed according to the optical power variation characteristics, and the coupling parameter vector includes six components of optical fiber end X, Y, Z displacement parameters and pitch angle, yaw angle and spacing parameters. The system limits the value range of each parameter to ±50 μm and ±1.5°, and quantizes them with a step size of 0.2 μm and 0.05°.

[0032] In the initialization stage, 40 bat individuals are generated by using the improved latent space reinforcement-bat algorithm. The coupling parameter vector of each individual is mapped to the latent space to form a potential feature vector by using the reversible flow coding. The initial loudness is set to 0.9, the pulse emission rate is set to 0.1, and the search frequency range is [0.5, 2.5]. In the evolution stage, the action output is calculated by using the latent space reinforcement strategy network module to realize the joint optimization of frequency adjustment, speed correction and position disturbance. The fitness value is calculated according to the optical power feedback signal after each iteration of the algorithm, and the strategy parameters are adaptively adjusted by using the reinforcement learning reward mechanism.

[0033] In order to verify the performance of the algorithm, the standard bat algorithm, the improved particle swarm algorithm and the improved latent space reinforcement-bat algorithm are selected for three groups of experiments. The experiments are carried out under the same environment, each group of experiment is independently run 20 times, and the maximum optical power value, the convergence iteration number, the optimization time and the optical power stability are taken as the main performance indicators.

[0034] Table 1 Comparison of optical fiber coupling performance based on different algorithms

[0035] As can be seen from Table 1, under the same experimental conditions, the improved latent space reinforcement-bat algorithm proposed in the present application is obviously superior to the traditional algorithm in the performance of optical fiber coupling optimization. In terms of optical power output, the maximum output power of the improved latent space reinforcement-bat algorithm reaches-1.42dBm, which is about 0.4dB higher than that of the standard bat algorithm and about 0.25dB higher than that of the improved particle swarm algorithm, indicating that the light field matching is more accurate and the coupling loss is smaller.

[0036] In terms of convergence performance, the average iteration number of the algorithm is 96, which is about 42% less than that of the standard bat algorithm and about 32% less than that of the improved particle swarm algorithm, which reflects faster convergence speed and better optimization efficiency, which is mainly due to the synergistic optimization mechanism of the latent space mapping and the reinforcement strategy network.

[0037] In terms of optimization efficiency, the average optimization time of the improved latent space reinforcement-bat algorithm is only 15.7 seconds, which is about 41% shorter than that of the standard bat algorithm, realizing fast and high-precision online parameter optimization. In terms of system stability, the optical power stability is reduced from 0.048 to 0.021, the output fluctuation is significantly reduced, and the algorithm's ability to suppress noise and drift is significantly enhanced.

[0038] In addition, the global optimal hit rate of the improved latent space reinforcement-bat algorithm in multiple experiments reaches 93.2%, which is significantly higher than that of the comparative algorithm, indicating that it has strong global search ability and result consistency. In summary, the method of the present application has achieved significant improvement in optical fiber coupling efficiency, convergence speed, optimization accuracy and system robustness, verifying its superiority and practical value in complex multi-parameter optical fiber coupling systems.

Claims

1. A method for improving fiber coupling efficiency based on a bat algorithm, characterized in that, The method comprises the following steps: Collecting optical power data of the fiber coupling system and preprocessing; Setting a coupling parameter vector and a value range of the fiber coupling system according to the preprocessed optical power data, and constructing a coupling parameter search space model; Initializing an improved latent space reinforcement-bat algorithm based on the coupling parameter search space model, and outputting an initial coupling parameter vector; Inputting the initial coupling parameter vector into a latent space coding module of the improved latent space reinforcement-bat algorithm, and introducing a reversible flow coding to generate a latent space vector; Inputting the latent space vector into a latent space reinforcement strategy network module of the improved latent space reinforcement-bat algorithm, and calculating an optimal action vector through a differentiable strategy flow mapping mechanism; Inputting the optimal action vector into a bat individual evolution update module of the improved latent space reinforcement-bat algorithm, and obtaining a current iteration optimal bat individual; Inputting the current iteration optimal bat individual into an adaptive adjustment and termination control module of the improved latent space reinforcement-bat algorithm, adaptively correcting the loudness and pulse emission rate of the current iteration optimal bat individual, adjusting the latent space reinforcement strategy parameters according to the feedback signal of the optical power data, and outputting a coupling parameter vector of a global optimal bat individual; According to the coupling parameter vector of the global optimal bat individual, the execution mechanism is controlled to adjust the fiber coupling system to the corresponding coupling parameter state and lock, and the optical power signal is continuously monitored during the operation of the fiber coupling system. When it is detected that the optical power signal is less than a preset threshold, the improved latent space reinforcement-bat algorithm is called again to perform parameter optimization calculation.

2. The method of claim 1, wherein the method is based on a bat algorithm. The preprocessing includes noise filtering, smoothing processing and normalization preprocessing.

3. The method of claim 1, wherein the method is based on a bat algorithm. The construction of the coupling parameter search space model specifically includes: Based on the preprocessed optical power signal, extracting optical power response features, the optical power response features including optical power amplitude change rate, coupling loss change trend and system steady-state power distribution; Defining a coupling parameter vector, and setting a lower limit and an upper limit for each parameter component according to the change interval of the optical power response features to form a lower limit vector and an upper limit vector, the coupling parameter vector including a fiber end position parameter, a fiber end angle parameter, and an optical component spacing parameter; Based on the lower limit vector and the upper limit vector, a coupling parameter feasible region is established, and a linear constraint matrix and a linear constraint vector are used to limit the linear dependence relationship between each coupling parameter component and the system feasible state to obtain a feasible region boundary model; Under the constraint condition of the feasible region boundary model, the parameter resolution step is set according to the dynamic change range of the preprocessed optical power data and the coupling parameter sensitivity, and the boundary information of the lower limit vector and the upper limit vector is used to quantize and normalize each component of the coupling parameter vector to obtain a quantized coupling parameter vector and a normalized coupling parameter vector; The lower limit vector, the upper limit vector, the linear constraint matrix, the linear constraint vector, the parameter resolution step, the quantized coupling parameter vector and the normalized coupling parameter vector are combined to form the coupling parameter search space model.

4. The method of claim 1, wherein the method is based on a bat algorithm. The output of the initial coupling parameter vector specifically includes: The improved latent space reinforcement-bat algorithm is loaded and initialized based on a coupling parameter search space model, and initialization parameters are set, including a population size, a maximum iteration number, a search frequency range, an initial loudness, and an initial pulse emission rate; A candidate coupling parameter vector is generated under the constraints of a linear constraint matrix and a linear constraint vector of the coupling parameter search space model based on the initialization parameters, and the generation of the candidate coupling parameter vector is based on a coupling parameter sample generated according to the initialization parameters, the coupling parameter sample is substituted into the constraint conditions defined by the linear constraint matrix and the linear constraint vector, the coupling parameter sample that does not satisfy the constraint conditions is modified, and a coupling parameter sample that satisfies all the constraint conditions is obtained as the candidate coupling parameter vector; For the candidate coupling parameter vector, an initial coupling parameter vector is generated for each bat individual within the limited range of the lower limit value vector and the upper limit value vector.

5. The method of claim 1, wherein the method is based on a bat algorithm. The generation of the latent space vector specifically includes: The initial coupling parameter vector is normalized to map the parameter values of the initial coupling parameter vector to a normalized interval of zero to one in proportion, and a normalized coupling parameter vector is generated; A coupling parameter correlation matrix is constructed based on the normalized coupling parameter vector, and a coupling parameter sensitivity coefficient is calculated, and the construction of the coupling parameter correlation matrix is a covariance analysis of the normalized values of each coupling parameter under multiple groups of optical power data, the linear correlation coefficient and the nonlinear coupling degree between each parameter component are calculated, and the coupling parameter correlation matrix is formed by row and column combination, and the calculation process of the coupling parameter sensitivity coefficient is that a small disturbance is applied to each component of the normalized coupling parameter vector in turn, the change of the optical power output before and after the disturbance is recorded, the ratio of the change of the optical power output to the corresponding parameter disturbance is compared, the local response degree of the component to the optical power output is obtained, and the ratio is defined as the coupling parameter sensitivity value of the component; The normalized coupling parameter vector and the coupling parameter sensitivity coefficient are input into the latent space encoding module of the improved latent space reinforcement-bat algorithm, the normalized coupling parameter vector is weighted and transformed through the feature weighting layer to obtain a weighted normalized coupling parameter vector; The weighted normalized coupling parameter vector is nonlinearly mapped by using the reversible flow coding to output the latent space vector, and the latent space vector refers to the state of the bat individual in the latent space feature domain.

6. The method of claim 1, wherein the method is based on a bat algorithm. The output of the optimal action vector specifically includes: The latent space vector is input into the latent space reinforcement strategy network module of the improved latent space reinforcement-bat algorithm, linear mapping and nonlinear activation calculation are performed through the differentiable policy flow mapping mechanism to obtain the latent space feature, and the latent space reinforcement strategy network module includes an input layer, a hidden layer, and an action generation layer; The latent space feature is input into the hidden layer of the latent space reinforcement strategy network module to perform feature mapping through a weight matrix and apply a nonlinear activation function to output an original action vector; The amplitude constraint processing is performed on the original action vector to obtain a standardized action output. The standardized action output is divided into a frequency adjustment amount, a speed correction amount, a position disturbance amount, a loudness adjustment amount and a pulse emission rate adjustment amount according to parameter categories, and each component is subjected to amplitude correction and constraint according to a preset scaling coefficient and a mapping rule, so as to obtain an optimal action vector.

7. The method of claim 1, wherein the method is based on a bat algorithm. The obtaining of the current iteration optimal bat individual specifically comprises: The search frequency of the bat individual is updated according to the frequency adjustment amount in the optimal action vector, so as to obtain an updated frequency adjustment amount, wherein the updated frequency adjustment amount is obtained by superimposing the search frequency and the frequency adjustment amount of the last iteration; The speed vector of the bat individual is updated according to the speed correction amount in the optimal action vector and the coupling parameter vector of the optimal bat individual, so as to obtain an updated speed vector, wherein the updated speed vector is obtained by adding the original speed vector and the speed correction amount, and the coupling parameter difference between the bat individual and the global optimal individual is weighted and corrected by combining the search frequency; The coupling parameter vector of the bat individual is updated according to the updated speed vector and the position disturbance amount in the optimal action vector, so as to obtain an updated position vector, wherein the updated position vector is obtained by the original position vector, the updated speed vector and the position disturbance amount; The updated frequency adjustment amount, the updated speed vector and the updated position vector are spliced to form an updated coupling parameter vector of the bat individual; The updated coupling parameter vector of the bat individual is input into the fiber coupling system, the corresponding optical power output signal is collected by controlling the actuator, and the fitness value of the bat individual is calculated, and the bat individual with the highest fitness value is selected as the optimal bat individual of the current iteration, wherein the calculation process of the fitness value of the bat individual is based on the intensity of the optical power output signal, the collected optical power signal is compared with the target maximum optical power, the minimum coupling loss is taken as the evaluation criterion, and the fitness value of each bat individual is calculated. 8.The method of claim 1, wherein, The output of the coupling parameter vector of the global optimal bat individual specifically comprises: The loudness adjustment amount and the pulse emission rate adjustment amount of the current iteration optimal bat individual are adaptively corrected to obtain a correction result, and the correction result is synchronously updated to the parameter state of the improved latent space reinforcement-bat algorithm; The strategy parameters of the latent space reinforcement strategy network module are adaptively updated through the updated parameter state of the improved latent space reinforcement-bat algorithm to form updated strategy parameters, and the updating process is performed by adjusting the strategy parameters in the direction of increasing reward value; According to the updated strategy parameters, a termination condition is determined, and the maximum iteration number is set as a preset upper limit, and when the iteration number reaches the preset upper limit, the coupling parameter vector of the global optimal bat individual is output. 9.The method of claim 1, wherein, The re-calling of the improved latent space reinforcement-bat algorithm specifically comprises: According to the coupling parameter vector of the global optimal bat individual, the coupling parameter state of the fiber coupling system is adjusted to the state corresponding to the global optimal coupling parameter vector by controlling the actuator, and a coupling parameter state locking operation is performed to form a global optimal coupling state. Based on the global optimal coupling state, set the online monitoring parameters of the optical power signal, collect the optical power signal according to the set sampling period, calculate the sliding average value of the optical power in each monitoring window, and the online monitoring parameters include the sampling period, the monitoring window length, the power threshold and the power drift threshold; When the sliding average value of the optical power is less than the preset power threshold, a re-optimization trigger signal is generated, and the generation of the re-optimization trigger signal is based on the optical power real-time monitoring module. When the sliding average optical power is lower than the preset power threshold in the continuous sampling period, the monitoring module determines that the system coupling state deviates from the optimal interval, and automatically outputs a digital trigger signal to the control unit; When the re-optimization trigger signal is detected to be in the active state, the coupling parameter state of the current optical fiber coupling system is read and input to the control unit together with the current optical power signal; In the control unit, an optimization start instruction is generated, the optimization process of the improved latent space reinforcement-bat algorithm is called again, and parameter re-optimization calculation is performed.

Images