A visible light communication channel selection optimization method based on genetic algorithm

Abstract

The present invention discloses a visible light communication channel selection optimization method based on genetic algorithm, aiming to improve the global optimization capability and dynamic adaptability of channel selection. The method comprises the following steps: high-precision parameter collection and preprocessing of available visible light communication channels to generate an initial population; designing a multi-objective fitness function, comprehensively considering channel capacity, interference, delay and spectrum utilization indicators, and evaluating each individual in the population; combining online learning and real-time feedback mechanisms to dynamically adjust the parameters of large-scale pre-trained language models and deep learning networks; adopting multi-objective optimization and local search mechanisms to achieve multi-objective optimization of channel selection; and finally outputting the optimal channel selection scheme. The method of the present invention has the advantages of strong global optimization capability, high dynamic adaptability, excellent communication quality and high system efficiency, and is suitable for complex and changeable communication environments.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a visible light communication channel selection optimization method based on a genetic algorithm. Background Art

[0002] With the continuous development of information technology and the popularization of the Internet, communication technology has also developed rapidly. Among the many communication technologies, visible light communication, as an emerging communication method, has received extensive attention and research due to its advantages such as rich spectrum resources, high security, and no electromagnetic interference. Visible light communication uses light sources in the visible light band for data transmission and is usually used in indoor high-speed wireless communication, smart lighting, and vehicle networking. However, with the complexity of application scenarios and the diversification of user needs, visible light communication faces many challenges in channel selection and optimization.

[0003] Existing visible light communication channel selection methods mainly rely on traditional optimization techniques, such as greedy algorithms, dynamic programming algorithms, etc. These methods can achieve certain results when dealing with channel selection problems in small-scale and simple scenarios. However, with the expansion of the scale of communication systems and the increase in scene complexity, the shortcomings of channel selection methods based on greedy algorithms in terms of global optimization capabilities and dynamic adaptability have gradually emerged. For example, although the greedy algorithm is simple to calculate, it is easy to fall into local optimality and it is difficult to find the global optimal solution; the dynamic programming algorithm is limited by its high computational complexity and limited scope of application, and it is difficult to deal with large-scale channel selection problems.

[0004] In addition, existing channel selection methods usually only consider a single or a few performance indicators, such as channel capacity, interference, delay, etc., ignoring the importance of multi-objective optimization. In practical applications, channel selection needs to comprehensively consider multiple performance indicators to achieve the optimization of overall performance. For example, there is a contradiction between channel capacity and interference. Simply optimizing channel capacity may lead to increased interference, which in turn affects communication quality. Therefore, how to achieve global optimization of channel selection under the framework of multi-objective optimization is an urgent problem to be solved.

[0005] Optimization methods based on genetic algorithms are widely used in complex optimization problems because of their global search capabilities and adaptive mechanisms. Genetic algorithms gradually approach the global optimal solution by simulating operations such as selection, crossover, and mutation in the biological evolution process. However, traditional genetic algorithms still face some challenges when dealing with high-dimensional, multi-objective optimization problems. First, genetic algorithms are prone to fall into local optimality during the search process, especially in complex multi-peak optimization problems, and it is necessary to introduce diversity maintenance mechanisms and global search strategies. Secondly, the parameter settings of genetic algorithms have a great impact on their performance. How to adaptively adjust these parameters to improve optimization efficiency and convergence speed is an important research direction.

[0006] Therefore, how to provide a visible light communication channel selection optimization method based on a genetic algorithm is a problem that those skilled in the art urgently need to solve. Summary of the invention

[0007] One purpose of the present invention is to propose a visible light communication channel selection optimization method based on genetic algorithm. The present invention significantly improves the global optimization capability, dynamic adaptability and multi-objective optimization effect of visible light communication channel selection by introducing a number of innovative optimization strategies and technologies, overcomes the shortcomings of the prior art, and has important theoretical value and practical application prospects.

[0008] A visible light communication channel selection optimization method based on a genetic algorithm according to an embodiment of the present invention comprises the following steps:

[0009] S1. Perform multi-level preprocessing on available visible light communication channels, including the collection of channel characteristic parameters, the construction of inter-channel interference model and preliminary channel availability assessment;

[0010] S2. Generate an initial population using a genetic algorithm, where each individual represents a channel selection scheme, and the gene code of the individual represents the specific allocation method of the channel;

[0011] S3. Design a multi-objective fitness function, comprehensively consider multiple performance indicators such as channel capacity, inter-channel interference, system delay, spectrum utilization and bit error rate, and use a dynamic weighting method to evaluate the fitness function;

[0012] S4. Fitness evaluation is performed based on a multi-objective evolutionary algorithm, individuals with higher fitness are selected to enter the next generation, and the improved tournament selection strategy and elite retention strategy are used to retain individuals;

[0013] S5. Perform adaptive multi-point crossover operation, adopt adaptive crossover rate and multi-point crossover strategy, dynamically adjust crossover parameters according to population evolution status, and introduce local search mechanism;

[0014] S6. Combine Gaussian mutation, non-uniform mutation and chaotic mutation strategies to perform multi-mode mutation operations;

[0015] S7. Repeat the fitness evaluation, selection, crossover and mutation operations through multiple generations of iterative evolution, and use an external archive to record and track non-dominated solutions;

[0016] S8. After reaching a preset convergence condition or number of iterations, output a channel selection solution.

[0017] Optionally, the S1 includes the following steps:

[0018] S11. Collect high-precision parameters of available visible light communication channels, and use high-precision optical power meters and spectrum analyzers to obtain the channel gain G of each channel. i 、Channel Interference I i and the channel noise N i ;

[0019] S12. Construct inter-channel interference model:

[0020]

[0021] Among them, C ij is the mutual interference coefficient between channel i and channel j, P i is the transmit power of channel i, G ij is the path gain from channel i to channel j, α is the signal attenuation coefficient, d ij is the distance between channel i and channel j;

[0022] S13. Based on the collected channel parameters and interference model, channel availability assessment is performed:

[0023]

[0024] Among them, ESNR i is the effective signal-to-noise ratio of channel i, β i is the signal correction factor of channel i, γ is the interference impact factor, the effective signal-to-noise ratio of each channel is calculated, and the available channels with effective signal-to-noise ratio higher than the preset threshold are screened out;

[0025] S14. Monitor the dynamic changes of the channel environment, set the monitoring period T, re-collect the channel parameters in each period, and update the channel gain G i (t), channel interference I i (t) and channel noise N i (t):

[0026] G i (t) = G i ·(1+△G i (t));

[0027] I i (t) = I i ·(1+△I i (t));

[0028] N i (t) = N i ·(1+△N i (t));

[0029] Among them, G i (t), I i(t) and N i (t) are the increment factors that change with time;

[0030] S15. Use nonlinear adaptive filtering technology to smooth the collected channel parameters, and use the extended Kalman filter to smooth the channel gain G i 、Channel Interference I i and channel noise N i To filter:

[0031] S16. Store the processed channel parameters in a database as input data for subsequent channel selection optimization.

[0032] Optionally, S2 includes the following steps:

[0033] S21. Generate an initial population using historical data and statistical characteristics, set the population size to N, each individual represents a channel selection scheme, and the gene code of each individual consists of M genes, representing the specific allocation method of M channels;

[0034] S22. Model the channel selection problem as a multidimensional combinatorial optimization problem and define the gene sequence of each individual:

[0035] X={x1,x2,…,x M};

[0036] Where X represents channel x i In the case of selection, the gene value is calculated according to the channel parameters and interference model through the following nonlinear mapping function:

[0037]

[0038] Among them, σ is a nonlinear activation function, ∈ is a small constant to avoid the denominator being zero;

[0039] S23. Adopt a heuristic initialization strategy based on reinforcement learning and use the high-quality channel selection scheme in historical data to generate some initial individuals:

[0040] X init =f RL (H);

[0041] Among them, H is the historical data set, f RL Generate functions for reinforcement learning-based heuristics;

[0042] S24. Generate the remaining initial individuals using random initialization strategy and distribution sampling technology:

[0043] X rand =sample(P,Θ);

[0044] Where P is the channel parameter set, Θ is the parameter distribution matrix, and sample is the random generation function based on distribution sampling;

[0045] S25. Introduce a diversity maintenance mechanism and use entropy increment to evaluate population diversity:

[0046]

[0047] Among them, p ij is the probability of the jth gene value of the i-th individual;

[0048] S26. Store the initialized population in the population library as initial data for subsequent fitness evaluation, selection, crossover and mutation operations.

[0049] Optionally, S3 includes the following steps:

[0050] S31. Design a multi-objective fitness function and evaluate the fitness of each individual. The fitness function integrates multiple performance indicators such as channel capacity C(X), channel interference I(X), system delay D(X), spectrum utilization U(X) and energy efficiency E(X);

[0051] S32. Define the channel capacity C(X) as a function of the total capacity of all channels:

[0052]

[0053] Among them, B i is the bandwidth of channel i, P i is the transmit power of channel i, G i is the gain of channel i, N i is the noise power of channel i, I i is the interference power of channel i;

[0054] S33. Define channel interference I(X) as a function of all inter-channel interference:

[0055]

[0056] Among them, C ij is the interference coefficient between channel i and channel j:

[0057]

[0058] Where α is the signal attenuation coefficient, d ij is the distance between channel i and channel j;

[0059] S34. Define the system delay D(X) as the delay function of the channel selection scheme:

[0060]

[0061] Among them, L i is the data load of channel i, C i is the capacity of channel i;

[0062] S35. Define the spectrum utilization U(X) as the spectrum efficiency function of the channel selection scheme:

[0063]

[0064] Among them, T i is the occupation time of channel i;

[0065] S36. Define energy efficiency E(X) as the energy efficiency function of the channel selection scheme:

[0066]

[0067] S37. Combining the above multiple performance indicators, design a multi-objective fitness function F(X):

[0068]

[0069] Among them, α, β, γ, δ and η are dynamically adjusted weight coefficients;

[0070] S38. Each individual in the population is evaluated by the fitness function F(X), its fitness value is calculated, and the individuals are sorted according to the fitness value, and the first k individuals with the highest fitness values ​​are selected as the initial elite individuals:

[0071] X elite =select_elite(X,k);

[0072] Among them, select_elite is the elite individual selection function, and k is the number of elite individuals.

[0073] Optionally, S4 includes the following steps:

[0074] S41. Based on the fitness value, select individuals with higher fitness to enter the next generation, and use the improved tournament selection strategy and elite retention strategy to retain individuals;

[0075] S42. In the tournament selection strategy, a dynamic tournament size adjustment mechanism is introduced to dynamically adjust the tournament size T according to the population fitness distribution:

[0076]

[0077] Among them, N is the population size, k is the adjustment factor, σ is the standard deviation of the current population fitness, and σ0 is the preset standard deviation threshold;

[0078] S43. In the elite retention strategy, the top r with the highest fitness value are selected. e Individuals are directly retained to the next generation:

[0079] X elite = retain_elite(X,r e );

[0080] Among them, retain_elite is the elite individual retention function;

[0081] S44. Select non-elite individuals and introduce a roulette wheel selection strategy based on fitness ranking:

[0082]

[0083] Among them, P i is the selection probability, F i is the fitness value of individual i, is the sum of all individual fitness values;

[0084] S45. To enhance population diversity, a selection strategy based on neighborhood search is added, and the neighborhood search radius r is defined. n , perform local selection within a neighborhood:

[0085] X neigh =neigh_select(X,r n );

[0086] Among them, neigh_select is the neighborhood search selection function;

[0087] S46. Introducing the Pareto frontier selection strategy in multi-objective optimization:

[0088]

[0089] Among them, d i is the dominance degree of individual i, 1(·) is the indicator function, when F i All objectives of F j 1 is 1 when , otherwise it is 0;

[0090] S47. For the selected individual, calculate its fitness contribution C i :

[0091]

[0092] Among them, △F i is the change in fitness value;

[0093] S48. Merge the selected individuals with the elite individuals to form a new population, provide basic data for subsequent crossover and mutation operations, and complete the optimization of the selection operation.

[0094] Optionally, the S7 includes the following steps:

[0095] S71. Combine online learning and real-time feedback mechanism to dynamically adjust the parameters of large-scale pre-trained language models and deep learning networks, and define the real-time feedback loss function L t :

[0096]

[0097] Among them, α and β are weight coefficients, L current is the loss at the current moment, L k is the historical loss of the past K moments;

[0098] S72. Update the parameters of a large-scale pre-trained language model through an online learning algorithm based on a gradient manifold:

[0099]

[0100] Among them, θ t is the model parameter at the tth moment, η is the learning rate, is the loss function L t For parameter θ t The gradient of , λ is the regularization coefficient, M(θ t ) is a regularization term based on manifold learning;

[0101] S73. Use real-time feedback mechanism to dynamically adjust the weights of deep learning network, introduce reward signal adjustment strategy based on reinforcement learning, and define reward signal R t :

[0102]

[0103] Among them, r i is the reward value of individual i, γ is the adjustment factor, F i is the fitness value of individual i, F avg is the average fitness value of the population;

[0104] S74. Dynamically adjust the learning rate of the deep learning network and define the adaptive learning rate η t for:

[0105]

[0106] Among them, η0 is the initial learning rate, λ is the adjustment factor, R t is the reward signal at the current moment, R avgis the historical average reward signal;

[0107] S75. Use the model performance monitoring mechanism based on graph neural network to build a dynamic graph of the channel selection optimization process:

[0108]

[0109] in, is the state of node i at time t+1, N(i) is the set of neighbor nodes of node i, c ij is the edge weight between node i and node j, W is the trainable weight matrix, and σ is the nonlinear activation function;

[0110] S76. Based on the output results of the graph neural network, the model performance is evaluated and adjusted, and the adaptive reinforcement learning method is used to complete the dynamic adjustment process of the model.

[0111] Optionally, the S8 includes the following steps:

[0112] S81. After reaching a preset number of iterations or fitness threshold, a final evaluation is performed on the individuals in the population;

[0113] S82. Apply the channel selection scheme to the visible light communication system, test and verify it in the actual environment, and collect actual operation data, including channel capacity, channel interference, system delay and spectrum utilization;

[0114] S83. Using the collected actual operation data, a real-time feedback mechanism is used to dynamically adjust the channel selection scheme, and a real-time adjustment function A is defined. t (X):

[0115] A t (X) = η·(F actual (X)-F pred (X));

[0116] Among them, η is the learning rate, F actual (X) is the fitness value in the actual environment, F pred (X) is the predicted fitness value;

[0117] S84. According to the result of the real-time feedback mechanism, the optimal channel selection scheme is iteratively optimized:

[0118] X t+1 =X t +ΔX t ;

[0119] Where ΔX t is the amount of adjustment based on real-time feedback;

[0120] S85. Finally output a dynamically optimized channel selection solution.

[0121] The beneficial effects of the present invention are:

[0122] 1. The present invention significantly enhances the global search capability of the genetic algorithm by introducing an adaptive multi-point crossover strategy and an adaptive mutation rate strategy, avoiding the defect that the traditional genetic algorithm is prone to fall into the local optimum. The multi-point crossover strategy ensures the diversity of the population and the breadth of the global search through a crossover method based on the channel interference matrix; the adaptive mutation rate strategy flexibly adjusts the mutation rate by combining multiple mutation methods such as Gaussian mutation, non-uniform mutation and chaotic mutation, thereby improving the exploration capability and convergence speed of the population.

[0123] 2. The present invention adopts a multi-objective optimization algorithm, which comprehensively considers multiple performance indicators such as channel capacity, interference, delay, and spectrum utilization in the channel selection process. The multi-objective optimization method can achieve a balance between multiple performance indicators and find the optimal channel selection scheme, thereby improving the overall performance and efficiency of the communication system.

[0124] 3. The present invention realizes dynamic adjustment and adaptive optimization in the process of channel selection optimization by combining online learning and real-time feedback mechanism. The online learning algorithm enables the model to adapt to the dynamically changing environment by updating the parameters of the large-scale pre-trained language model and the deep learning network in real time; the real-time feedback mechanism dynamically adjusts the weights and learning rate of the deep learning network through the reward signal adjustment strategy based on reinforcement learning, further improving the adaptability and efficiency of channel selection optimization.

[0125] 4. The present invention optimizes the data retrieval process by adopting cache strategies and index optimization techniques, significantly improving retrieval efficiency and cache hit rate. The hash map-based index structure and multi-level index strategy ensure the efficiency and accuracy of index queries; the incremental update mechanism and the priority queue-based cache replacement strategy ensure the dynamic maintenance and efficient update of cache and index. BRIEF DESCRIPTION OF THE DRAWINGS

[0126] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0127] Figure 1 A flowchart of a visible light communication channel selection optimization method based on a genetic algorithm proposed by the present invention; DETAILED DESCRIPTION

[0128] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0129] refer to Figure 1 , a visible light communication channel selection optimization method based on genetic algorithm, comprising the following steps:

[0130] S1. Perform multi-level preprocessing on available visible light communication channels, including the collection of channel characteristic parameters, the construction of inter-channel interference model and preliminary channel availability assessment;

[0131] S2. Generate an initial population using a genetic algorithm, where each individual represents a channel selection scheme, and the gene code of the individual represents the specific allocation method of the channel;

[0132] S3. Design a multi-objective fitness function, comprehensively consider multiple performance indicators such as channel capacity, inter-channel interference, system delay, spectrum utilization and bit error rate, and use a dynamic weighting method to evaluate the fitness function;

[0133] S4. Fitness evaluation is performed based on a multi-objective evolutionary algorithm, individuals with higher fitness are selected to enter the next generation, and the improved tournament selection strategy and elite retention strategy are used to retain individuals;

[0134] S5. Perform adaptive multi-point crossover operation, adopt adaptive crossover rate and multi-point crossover strategy, dynamically adjust crossover parameters according to population evolution status, and introduce local search mechanism;

[0135] S6. Combine Gaussian mutation, non-uniform mutation and chaotic mutation strategies to perform multi-mode mutation operations;

[0136] S7. Repeat the fitness evaluation, selection, crossover and mutation operations through multiple generations of iterative evolution, and use an external archive to record and track non-dominated solutions;

[0137] S8. After reaching a preset convergence condition or number of iterations, output a channel selection solution.

[0138] In this implementation, S1 includes the following steps:

[0139] S11. Collect high-precision parameters of available visible light communication channels, and use high-precision optical power meters and spectrum analyzers to obtain the channel gain G of each channel. i 、Channel Interference I i and the channel noise N i ;

[0140] S12. Construct inter-channel interference model:

[0141]

[0142] Among them, C ij is the mutual interference coefficient between channel i and channel j, P i is the transmit power of channel i, G ij is the path gain from channel i to channel j, α is the signal attenuation coefficient, d ij is the distance between channel i and channel j;

[0143] S13. Based on the collected channel parameters and interference model, channel availability assessment is performed:

[0144]

[0145] Among them, ESNR i is the effective signal-to-noise ratio of channel i, β i is the signal correction factor of channel i, γ is the interference impact factor, the effective signal-to-noise ratio of each channel is calculated, and the available channels with effective signal-to-noise ratio higher than the preset threshold are screened out;

[0146] S14. Monitor the dynamic changes of the channel environment, set the monitoring period T, re-collect the channel parameters in each period, and update the channel gain G i (t), channel interference I i (t) and channel noise N i (t):

[0147] G i (t) = G i ·(1+△G i (t));

[0148] I i (t) = I i ·(1+△I i (t));

[0149] N i (t) = N i ·(1+△N i (t));

[0150] Among them, G i (t), I i (t) and N i (t) are the increment factors that change with time;

[0151] S15. Use nonlinear adaptive filtering technology to smooth the collected channel parameters, and use the extended Kalman filter to smooth the channel gain G i 、Channel Interference I i and channel noise N i To filter:

[0152] S16. Store the processed channel parameters in a database as input data for subsequent channel selection optimization.

[0153] In this implementation, S2 includes the following steps:

[0154] S21. Generate an initial population using historical data and statistical characteristics, set the population size to N, each individual represents a channel selection scheme, and the gene code of each individual consists of M genes, representing the specific allocation method of M channels;

[0155] S22. Model the channel selection problem as a multidimensional combinatorial optimization problem and define the gene sequence of each individual:

[0156] X={x1,x2,…,x M};

[0157] Where X represents channel x i In the case of selection, the gene value is calculated according to the channel parameters and interference model through the following nonlinear mapping function:

[0158]

[0159] Among them, σ is a nonlinear activation function, ∈ is a small constant to avoid the denominator being zero;

[0160] S23. Adopt a heuristic initialization strategy based on reinforcement learning and use the high-quality channel selection scheme in historical data to generate some initial individuals:

[0161] X init =f RL (H);

[0162] Among them, H is the historical data set, f RL Generate functions for reinforcement learning-based heuristics;

[0163] S24. Generate the remaining initial individuals using random initialization strategy and distribution sampling technology:

[0164] X rand =sample(P,Θ);

[0165] Where P is the channel parameter set, Θ is the parameter distribution matrix, and sample is the random generation function based on distribution sampling;

[0166] S25. Introduce a diversity maintenance mechanism and use entropy increment to evaluate population diversity:

[0167]

[0168] Among them, p ijis the probability of the jth gene value of the i-th individual;

[0169] S26. Store the initialized population in the population library as initial data for subsequent fitness evaluation, selection, crossover and mutation operations.

[0170] In this implementation, S3 includes the following steps:

[0171] S31. Design a multi-objective fitness function and evaluate the fitness of each individual. The fitness function integrates multiple performance indicators such as channel capacity C(X), channel interference I(X), system delay D(X), spectrum utilization U(X) and energy efficiency E(X);

[0172] S32. Define the channel capacity C(X) as a function of the total capacity of all channels:

[0173]

[0174] Among them, B i is the bandwidth of channel i, P i is the transmit power of channel i, G i is the gain of channel i, N i is the noise power of channel i, I i is the interference power of channel i;

[0175] S33. Define channel interference I(X) as a function of all inter-channel interference:

[0176]

[0177] Among them, C ij is the interference coefficient between channel i and channel j:

[0178]

[0179] Where α is the signal attenuation coefficient, d ij is the distance between channel i and channel j;

[0180] S34. Define the system delay D(X) as the delay function of the channel selection scheme:

[0181]

[0182] Among them, L i is the data load of channel i, C i is the capacity of channel i;

[0183] S35. Define the spectrum utilization U(X) as the spectrum efficiency function of the channel selection scheme:

[0184]

[0185] Among them, T i is the occupation time of channel i;

[0186] S36. Define energy efficiency E(X) as the energy efficiency function of the channel selection scheme:

[0187]

[0188] S37. Combining the above multiple performance indicators, design a multi-objective fitness function F(X):

[0189]

[0190] Among them, α, β, γ, δ and η are dynamically adjusted weight coefficients;

[0191] S38. Each individual in the population is evaluated by the fitness function F(X), its fitness value is calculated, and the individuals are sorted according to the fitness value, and the first k individuals with the highest fitness values ​​are selected as the initial elite individuals:

[0192] X elite =select_elite(X,k);

[0193] Among them, select_elite is the elite individual selection function, and k is the number of elite individuals.

[0194] In this implementation manner, S4 includes the following steps:

[0195] S41. Based on the fitness value, select individuals with higher fitness to enter the next generation, and use the improved tournament selection strategy and elite retention strategy to retain individuals;

[0196] S42. In the tournament selection strategy, a dynamic tournament size adjustment mechanism is introduced to dynamically adjust the tournament size T according to the population fitness distribution:

[0197]

[0198] Among them, N is the population size, k is the adjustment factor, σ is the standard deviation of the current population fitness, and σ0 is the preset standard deviation threshold;

[0199] S43. In the elite retention strategy, the top r with the highest fitness value are selected. e Individuals are directly retained to the next generation:

[0200] X elite = retain_elite(X,r e );

[0201] Among them, retain_elite is the elite individual retention function;

[0202] S44. Select non-elite individuals and introduce a roulette wheel selection strategy based on fitness ranking:

[0203]

[0204] Among them, P i is the selection probability, F i is the fitness value of individual i, is the sum of all individual fitness values;

[0205] S45. To enhance population diversity, a selection strategy based on neighborhood search is added, and the neighborhood search radius r is defined. n , perform local selection within a neighborhood:

[0206] X neigh =neigh_select(X,r n );

[0207] Among them, neigh_select is the neighborhood search selection function;

[0208] S46. Introducing the Pareto frontier selection strategy in multi-objective optimization:

[0209]

[0210] Among them, d i is the dominance degree of individual i, 1(·) is the indicator function, when F i All objectives of F j 1 is 1 when , otherwise it is 0;

[0211] S47. For the selected individual, calculate its fitness contribution C i :

[0212]

[0213] Among them, △F i is the change in fitness value;

[0214] S48. Merge the selected individuals with the elite individuals to form a new population, provide basic data for subsequent crossover and mutation operations, and complete the optimization of the selection operation.

[0215] In this implementation manner, the S7 includes the following steps:

[0216] S71. Combine online learning and real-time feedback mechanism to dynamically adjust the parameters of large-scale pre-trained language models and deep learning networks, and define the real-time feedback loss function Lt :

[0217]

[0218] Among them, α and β are weight coefficients, L current is the loss at the current moment, L k is the historical loss of the past K moments;

[0219] S72. Update the parameters of a large-scale pre-trained language model through an online learning algorithm based on a gradient manifold:

[0220]

[0221] Among them, θ t is the model parameter at the tth moment, η is the learning rate, is the loss function L t For parameter θ t The gradient of , λ is the regularization coefficient, M(θ t ) is a regularization term based on manifold learning;

[0222] S73. Use real-time feedback mechanism to dynamically adjust the weights of deep learning network, introduce reward signal adjustment strategy based on reinforcement learning, and define reward signal R t :

[0223]

[0224] Among them, r i is the reward value of individual i, γ is the adjustment factor, F i is the fitness value of individual i, F avg is the average fitness value of the population;

[0225] S74. Dynamically adjust the learning rate of the deep learning network and define the adaptive learning rate η t for:

[0226]

[0227] Among them, η0 is the initial learning rate, λ is the adjustment factor, R t is the reward signal at the current moment, R avg is the historical average reward signal;

[0228] S75. Use the model performance monitoring mechanism based on graph neural network to build a dynamic graph of the channel selection optimization process:

[0229]

[0230] in, is the state of node i at time t+1, N(i) is the set of neighbor nodes of node i, c ij is the edge weight between node i and node j, W is the trainable weight matrix, and σ is the nonlinear activation function;

[0231] S76. Based on the output results of the graph neural network, the model performance is evaluated and adjusted, and the adaptive reinforcement learning method is used to complete the dynamic adjustment process of the model.

[0232] In this implementation manner, the S8 includes the following steps:

[0233] S81. After reaching a preset number of iterations or fitness threshold, a final evaluation is performed on the individuals in the population;

[0234] S82. Apply the channel selection scheme to the visible light communication system, test and verify it in the actual environment, and collect actual operation data, including channel capacity, channel interference, system delay and spectrum utilization;

[0235] S83. Using the collected actual operation data, a real-time feedback mechanism is used to dynamically adjust the channel selection scheme, and a real-time adjustment function A is defined. t (X):

[0236] A t (X) = η·(F actual (X)-F pred (X));

[0237] Among them, η is the learning rate, F actual (X) is the fitness value in the actual environment, F pred (X) is the predicted fitness value;

[0238] S84. According to the result of the real-time feedback mechanism, the optimal channel selection scheme is iteratively optimized:

[0239] X t+1 =X t +ΔX t ;

[0240] Where ΔX t is the amount of adjustment based on real-time feedback;

[0241] S85. Finally output a dynamically optimized channel selection solution.

[0242] Embodiment 1:

[0243] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the indoor communication environment of a smart city, and a high-speed wireless communication system based on visible light communication is deployed to meet the high data transmission requirements in a high-density user environment. The system adopts the channel selection optimization method based on genetic algorithm proposed by us, aiming to solve the problems of the channel selection method based on greedy algorithm that is prone to fall into local optimality, cannot dynamically adapt to environmental changes, and lacks multi-objective optimization. The test location is located in a smart office building in the city center in June 2024, and the test data collection period is four weeks.

[0244] In the communication system deployment of this office building, light source equipment is installed on the ceiling, and each office and conference room is equipped with LED lamps, which not only provide lighting but also undertake the function of data transmission. To ensure the quality of communication, the LED lamps in each room are connected to a centralized control system, which monitors the usage of optical communication channels, inter-channel interference, and data transmission quality in real time. Our optimization method is applied in this environment to dynamically select and optimize communication channels and improve the overall system performance.

[0245] When the system is initialized, high-precision parameters of the visible light communication channel are collected, including channel gain, channel interference, and channel noise. The basic characteristic parameters of each channel are obtained through a high-precision optical power meter and spectrum analyzer. Then, the initial population is generated using historical data and statistical characteristics, and each individual represents a channel selection scheme. The individual gene code is generated through a heuristic initialization strategy and a random initialization strategy to ensure the diversity of the population.

[0246] In actual application, the system uses a multi-objective fitness function to evaluate the fitness of each individual, taking into account multiple performance indicators such as channel capacity, interference, delay and spectrum utilization. The design of the fitness function can achieve a balance between multiple performance indicators and find the optimal channel selection scheme. For the channel selection problem, the genetic algorithm adopts an adaptive multi-point crossover strategy. Through a multi-point crossover method based on the channel interference matrix, the crossover point and crossover probability are determined, thereby enhancing the diversity and global search ability of the population. The mutation operation adopts an adaptive mutation rate strategy, combining multiple mutation methods such as Gaussian mutation, non-uniform mutation and chaotic mutation, which improves the exploration ability and convergence speed of the population.

[0247] To further optimize the channel selection process, the system combines online learning and real-time feedback mechanisms to dynamically adjust the parameters of large-scale pre-trained language models and deep learning networks. By updating model parameters in real time and adjusting reward signal strategies based on reinforcement learning, the system can adapt to dynamically changing environments and ensure the optimality and stability of channel selection. In addition, the system also adopts a model performance monitoring mechanism based on graph neural networks. By constructing a dynamic graph during the channel selection optimization process, the system evaluates and adjusts the model performance in real time to ensure efficient and stable performance in complex and changing environments.

[0248] During the four-week test period, the system ran 24 hours a day and recorded data every hour, including key performance indicators such as channel capacity, channel interference, system latency, spectrum utilization, and energy efficiency. By comparing the traditional channel selection method with our proposed method, the test data showed significant performance improvement, as shown in Table 1 below:

[0249] Table 1: Weekly average data comparison table

[0250]

[0251]

[0252] In the first week of testing, the greedy algorithm-based channel selection method had an average channel capacity of 600Mbps, channel interference of 25%, system latency of 50ms, spectrum utilization of 70%, and energy efficiency of 85%. With our method, the channel capacity increased to 850Mbps, channel interference decreased to 15%, system latency decreased to 30ms, spectrum utilization increased to 85%, and energy efficiency increased to 92%.

[0253] During the peak period of the second week, the user density in the office building increased, and the channel capacity of the greedy algorithm-based channel selection method dropped to 550Mbps, the channel interference rose to 30%, the system delay increased to 60ms, the spectrum utilization rate dropped to 65%, and the energy efficiency dropped to 80%. With our method, the channel capacity remained at 820Mbps, the channel interference was only 18%, the system delay remained at 35ms, the spectrum utilization rate was stable at 82%, and the energy efficiency was 90%.

[0254] During the system maintenance in the third week, the system adaptively adjusted the parameters of the large-scale pre-trained language model and deep learning network through online learning and real-time feedback mechanism. The channel capacity of the channel selection method based on the greedy algorithm fluctuates greatly, while our method can quickly adapt to environmental changes, the channel capacity is stable at 830Mbps, the channel interference is controlled at 16%, the system delay is 32ms, and the spectrum utilization and energy efficiency are 84% and 91% respectively.

[0255] In the last week, there were several large meetings and events in the office building, and the test data further proved the superiority of our method. The channel capacity of the greedy algorithm-based channel selection method dropped again to 520Mbps, the channel interference reached 35%, the system delay increased to 65ms, the spectrum utilization rate dropped to 60%, and the energy efficiency was only 78%. Our method maintained good performance, with a channel capacity of 810Mbps, a channel interference of 20%, a system delay of 38ms, and a spectrum utilization and energy efficiency of 80% and 88% respectively.

[0256] Through four weeks of practical application and data collection, the visible light communication channel selection optimization method based on genetic algorithm proposed by us is significantly superior to the channel selection method based on greedy algorithm in terms of improving channel capacity, reducing channel interference, reducing system delay, improving spectrum utilization and energy efficiency. These data prove the effectiveness and superiority of our method, demonstrate its strong adaptability and stability in high-density user environments, and provide reliable technical support and solutions for future visible light communication systems.

[0257] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A visible light communication channel selection optimization method based on genetic algorithm, characterized in that: The following steps are involved: S1. Perform multi-level preprocessing on available visible light communication channels, including the collection of channel characteristic parameters, the construction of inter-channel interference model and preliminary channel availability assessment; S2. Generate an initial population using a genetic algorithm, where each individual represents a channel selection scheme, and the gene code of the individual represents the specific allocation method of the channel; S3. Design a multi-objective fitness function, comprehensively consider multiple performance indicators such as channel capacity, inter-channel interference, system delay, spectrum utilization and energy efficiency, and use a dynamic weighting method to evaluate the fitness function; S4. Fitness evaluation is performed based on a multi-objective evolutionary algorithm, individuals with higher fitness are selected to enter the next generation, and the improved tournament selection strategy and elite retention strategy are used to retain individuals; S5. Perform adaptive multi-point crossover operation, adopt adaptive crossover rate and multi-point crossover strategy, dynamically adjust crossover parameters according to population evolution status, and introduce local search mechanism; S6. Combine Gaussian mutation, non-uniform mutation and chaotic mutation to perform multi-mode mutation operation; S7. Repeat the fitness evaluation, selection, crossover and mutation operations through multiple generations of iterative evolution, and use an external archive to record and track non-dominated solutions; S8. After reaching the preset convergence condition or number of iterations, output the channel selection scheme; S31. Design a multi-objective fitness function and evaluate the fitness of each individual. The fitness function integrates multiple performance indicators such as channel capacity C(X), channel interference I(X), system delay D(X), spectrum utilization U(X) and energy efficiency E(X); S32. Define the channel capacity C(X) as a function of the total capacity of all channels: Among them, B i is the bandwidth of channel i, P i is the transmit power of channel i, G i is the gain of channel i, N i is the noise power of channel i, I i is the interference power of channel i; S33. Define channel interference I(X) as a function of all inter-channel interference: Among them, C ij is the interference coefficient between channel i and channel j: Where α is the signal attenuation coefficient, d ij is the distance between channel i and channel j; S34. Define the system delay D(X) as the delay function of the channel selection scheme: Among them, L i is the data load of channel i, C i is the capacity of channel i; S35. Define the spectrum utilization U(X) as the spectrum efficiency function of the channel selection scheme: Among them, T i is the occupation time of channel i; S36. Define energy efficiency E(X) as the energy efficiency function of the channel selection scheme: S37. Combining the above multiple performance indicators, design a multi-objective fitness function F(X): Among them, α, β, γ, δ and η are dynamically adjusted weight coefficients; S71. Combine online learning and real-time feedback mechanism to dynamically adjust the parameters of large-scale pre-trained language models and deep learning networks, and define the real-time feedback loss function L t : Among them, α and β are weight coefficients, L current is the loss at the current moment, L k is the historical loss of the past K moments; S72. Update the parameters of a large-scale pre-trained language model through an online learning algorithm based on a gradient manifold: Among them, θ t is the model parameter at the tth moment, η is the learning rate, is the loss function L t For parameter θ t The gradient of , λ is the regularization coefficient, M(θ t ) is a regularization term based on manifold learning; S73. Use real-time feedback mechanism to dynamically adjust the weights of deep learning network, introduce reward signal adjustment strategy based on reinforcement learning, and define reward signal R t : Among them, r i is the reward value of individual i, γ is the adjustment factor, F i is the fitness value of individual i, F avg is the average fitness value of the population; S74. Dynamically adjust the learning rate of the deep learning network and define the adaptive learning rate η t for: Among them, η0 is the initial learning rate, λ is the adjustment factor, R t is the reward signal at the current moment, R avg is the historical average reward signal.

2. According to claim 1, a visible light communication channel selection optimization method based on genetic algorithm is characterized in that: The S1 specifically includes: S11. Collect high-precision parameters of available visible light communication channels, and use high-precision optical power meters and spectrum analyzers to obtain the channel gain G of each channel. i 、Channel Interference I i and the channel noise N i ; S12. Construct inter-channel interference model: Among them, C ij is the mutual interference coefficient between channel i and channel j, P i is the transmit power of channel i, G ij is the path gain from channel i to channel j, α is the signal attenuation coefficient, d ij is the distance between channel i and channel j; S13. Based on the collected channel parameters and interference model, channel availability assessment is performed: Among them, ESNR i is the effective signal-to-noise ratio of channel i, β i is the signal correction factor of channel i, γ is the interference impact factor, the effective signal-to-noise ratio of each channel is calculated, and the available channels with effective signal-to-noise ratio higher than the preset threshold are screened out; S14. Monitor the dynamic changes of the channel environment, set the monitoring period T, re-collect the channel parameters in each period, and update the channel gain G i (t), channel interference I i (t) and channel noise N i (t): G i (t)=G i ·(1+△G i (t)); I i (t)=I i ·(1+△I i (t)); N i (t)=N i ·(1+△N i (t)); Among them, G i (t), I i (t) and N i (t) are the increment factors that change with time; S15. Use nonlinear adaptive filtering technology to smooth the collected channel parameters, and use the extended Kalman filter to smooth the channel gain G i 、Channel Interference I i and channel noise N i To filter: S16. Store the processed channel parameters in a database as input data for subsequent channel selection optimization.

3. The visible light communication channel selection optimization method based on genetic algorithm according to claim 1, characterized in that: The S2 specifically includes: S21. Generate an initial population using historical data and statistical characteristics, set the population size to N, each individual represents a channel selection scheme, and the gene code of each individual consists of M genes, representing the specific allocation method of M channels; S22. Model the channel selection problem as a multidimensional combinatorial optimization problem and define the gene sequence of each individual: X={x1,x2,…,x M }; Where X represents channel x i In the case of selection, the gene value is calculated according to the channel parameters and interference model through the following nonlinear mapping function: Among them, σ is a nonlinear activation function, ∈ is a small constant to avoid the denominator being zero; S23. Adopt a heuristic initialization strategy based on reinforcement learning and use the high-quality channel selection scheme in historical data to generate some initial individuals: X init =f RL (H); Among them, H is the historical data set, f RL Generate functions for reinforcement learning-based heuristics; S24. Generate the remaining initial individuals using random initialization strategy and distribution sampling technology: X rand =sample(P,Θ); Where P is the channel parameter set, Θ is the parameter distribution matrix, and sample is the random generation function based on distribution sampling; S25. Introduce a diversity maintenance mechanism and use entropy increment to evaluate population diversity: Among them, p ij is the probability of the jth gene value of the i-th individual; S26. Store the initialized population in the population library as initial data for subsequent fitness evaluation, selection, crossover and mutation operations.

4. The visible light communication channel selection optimization method based on genetic algorithm according to claim 1, characterized in that: The S3 further includes: S38. Each individual in the population is evaluated by the fitness function F(X), its fitness value is calculated, and the individuals are sorted according to the fitness value, and the first k individuals with the highest fitness values ​​are selected as the initial elite individuals: X elite =select_elite(X,k); Among them, select_elite is the elite individual selection function, and k is the number of elite individuals.

5. The visible light communication channel selection optimization method based on genetic algorithm according to claim 1, characterized in that: The S4 specifically includes: S41. Based on the fitness value, select individuals with higher fitness to enter the next generation, and use the improved tournament selection strategy and elite retention strategy to retain individuals; S42. In the tournament selection strategy, a dynamic tournament size adjustment mechanism is introduced to dynamically adjust the tournament size T according to the population fitness distribution: Among them, N is the population size, k is the adjustment factor, σ is the standard deviation of the current population fitness, and σ0 is the preset standard deviation threshold; S43. In the elite retention strategy, the top r with the highest fitness value are selected. e Individuals are directly retained to the next generation: X elite =retain_elite(X,r e ); Among them, retain_elite is the elite individual retention function; S44. Select non-elite individuals and introduce a roulette wheel selection strategy based on fitness ranking: Among them, P i is the selection probability, F i is the fitness value of individual i, is the sum of all individual fitness values; S45. To enhance population diversity, a selection strategy based on neighborhood search is added, and the neighborhood search radius r is defined. n , perform local selection within a neighborhood: X neigh =neigh_select(X,r n ); Among them, neigh_select is the neighborhood search selection function; S46. Introducing the Pareto frontier selection strategy in multi-objective optimization: Among them, d i is the dominance degree of individual i, 1(·) is the indicator function, when F i All objectives of j 1 is 1 when , otherwise it is 0; S47. For the selected individual, calculate its fitness contribution C i : Among them, △F i is the change in fitness value; S48. Merge the selected individuals with the elite individuals to form a new population, provide basic data for subsequent crossover and mutation operations, and complete the optimization of the selection operation.

6. The visible light communication channel selection optimization method based on genetic algorithm according to claim 2, characterized in that: The S7 further comprises: S75. Use the model performance monitoring mechanism based on graph neural network to build a dynamic graph of the channel selection optimization process: in, is the state of node i at time t+1, N(i) is the set of neighbor nodes of node i, c ij is the edge weight between node i and node j, W is the trainable weight matrix, and σ is the nonlinear activation function; S76. Based on the output results of the graph neural network, the model performance is evaluated and adjusted, and the adaptive reinforcement learning method is used to complete the dynamic adjustment process of the model.

7. The visible light communication channel selection optimization method based on genetic algorithm according to claim 1, characterized in that: The S8 specifically includes: S81. After reaching a preset number of iterations or fitness threshold, a final evaluation is performed on the individuals in the population; S82. Apply the channel selection scheme to the visible light communication system, test and verify it in the actual environment, and collect actual operation data, including channel capacity, channel interference, system delay and spectrum utilization; S83. Using the collected actual operation data, a real-time feedback mechanism is used to dynamically adjust the channel selection scheme, and a real-time adjustment function A is defined. t (X): A t (X)=η·(F actual (X)-F pred (X)); Among them, η is the learning rate, F actual (X) is the fitness value in the actual environment, F pred (X) is the predicted fitness value; S84. According to the result of the real-time feedback mechanism, the optimal channel selection scheme is iteratively optimized: X t+1 =X t +ΔX t ; Where ΔX t is the amount of adjustment based on real-time feedback; S85. Finally output a dynamically optimized channel selection solution.

Images