MZI matrix self-optimization method and device based on genetic algorithm

Abstract

The invention discloses an MZI matrix self-optimization method and device based on a genetic algorithm. The method comprises the following steps: S1, initializing a target matrix; s2, designing an MZI network layout; s3, generating an initial population; s4, defining a fitness function, and calculating a fitness value according to parameter configuration of the phase shifter; s5, selecting individuals with high fitness by adopting a roulette method; s6, performing crossover operation on the selected individuals to generate a new generation of individuals, and updating the population; s7, mutation operation is carried out, varied individuals are brought into the current population, and the population is updated; and S8, repeatedly executing the steps S4-S7, judging whether a stop condition is met or not according to the convergence condition, and if the stop condition is met, outputting an optimal solution. The MZI network optimization method based on the genetic algorithm is adopted, self-optimization of the MZI matrix is achieved by optimizing parameter configuration of the phase shifter, and the method has the advantages of being efficient, accurate and capable of achieving global search.

Description

Technical Field

[0001] The present invention relates to the field of MZI matrix self-optimization, and in particular to a method and device for MZI matrix self-optimization based on a genetic algorithm. Background Art

[0002] In recent years, with the rapid development of information technology, photonic computing, as an emerging computing method, has received more and more attention. Compared with traditional electronic computing, photonic computing has higher parallel processing capabilities, lower power consumption and faster data transmission speed, so it has broad application prospects in high-performance computing, quantum computing, optical communications and other fields. Among the many photonic computing technologies, the matrix transformation method based on MZI network is considered to be an important implementation method.

[0003] The MZI network is a structure that uses the interference effect of light waves for phase modulation and optical transformation, and is usually composed of multiple MZI arrays. By adjusting the phase shifter voltage in each MZI, the MZI network can control the input optical signal and complete the matrix transformation. The MZI network has been widely used in optical signal processing, optical communication, optical computing and other fields due to its simple structure, flexible adjustment and fast response speed. Using the MZI network for matrix transformation can not only achieve efficient signal processing, but also greatly improve the computing power of the photonic computing system.

[0004] However, the existing matrix transformation methods based on MZI networks still have some limitations, especially in complex high-dimensional matrix operations and the processing of non-ideal parameters. First, the number and structural configuration of the phase shifters in the MZI network have a crucial impact on the accuracy and effect of the matrix transformation. As the scale of the MZI array increases, the complexity of the network increases dramatically, which brings great challenges to the optimization and design of the network. Traditional optimization methods often rely on manual adjustment or parameter optimization based on classical algorithms, but these methods are inefficient in the context of high-dimensional matrix transformation, are prone to fall into local optimal solutions, and cannot cope with complex system errors and uncertainties.

[0005] Secondly, existing optimization algorithms often lack sufficient global search capabilities and diversity when solving high-dimensional optical transformation problems. Especially in complex optimization tasks that require a lot of calculations and weigh multiple objectives, existing algorithms are easily affected by the initial parameters, resulting in unsatisfactory optimization results. For example, in practical applications, the parameters of the phase shifter in the MZI network, such as the voltage control value, are often affected by environmental noise and system errors, resulting in reduced accuracy of optical transformation. If we only rely on traditional gradient descent or local optimization methods, these errors may not be effectively overcome, ultimately resulting in reduced transformation accuracy or slow optimization speed.

[0006] In order to meet these challenges, genetic algorithms, as a global optimization algorithm based on natural selection and genetics principles, have shown excellent performance in complex optimization problems. Genetic algorithms have strong global search capabilities and high solution diversity, and can effectively avoid falling into local optimal solutions. Therefore, they are widely used in optimization tasks in various fields. Genetic algorithms gradually optimize individuals in the population by simulating the selection, crossover, mutation and other mechanisms in the biological evolution process in nature, thereby obtaining the optimal solution. In the field of optical computing, the introduction of genetic algorithms can significantly improve the optimization efficiency of MZI networks, especially in high-dimensional matrix operations and complex optical transformation tasks, where genetic algorithms can give full play to their advantages.

[0007] Although the application of genetic algorithms in MZI network optimization has potential, existing technologies still have some defects. First, when implementing MZI network optimization, existing genetic algorithm optimization methods often do not fully consider the mutual influence between the various phase shifters in the network and the change of optical coherence. The voltage control value of each phase shifter in the MZI network not only affects the optical transformation effect of the phase shifter itself, but also affects the optical interference result of the entire network. If these mutual influences are not considered, the optimization results may be unstable or the required optical transformation cannot be effectively achieved.

[0008] Secondly, although genetic algorithms have strong global search capabilities, their convergence speed and optimization accuracy are still affected by the initial population configuration, selection operations, crossover operations, and mutation operations. In the optimization process of MZI networks, how to reasonably select fitness functions, control crossover rates and mutation rates, and how to set appropriate population sizes and stop conditions are all key factors affecting algorithm performance. Traditional genetic algorithms often rely on experience in these parameter settings, and lack systematic optimization strategies, which makes it difficult to guarantee the effectiveness of the algorithm.

[0009] In addition, the computational complexity of genetic algorithms is high, especially when multiple iterations and a large number of population calculations are required, which may lead to an increase in computing time. This may become a performance bottleneck for optical computing systems that require real-time response. In the real-time optimization of MZI networks, how to balance the optimization effect and computational efficiency is an important issue in the current application of genetic algorithms. Summary of the invention

[0010] One object of the present invention is to propose a method for self-optimization of an MZI matrix based on a genetic algorithm. The present invention designs a fitness function, fully considers the interaction and coherence between each phase shifter in the MZI network, and adopts an optimization strategy to achieve efficient optimization of the MZI network, and improves the global search capability and convergence speed of the algorithm, so that the optimization result can be more accurate and stable.

[0011] A method for self-optimization of an MZI matrix based on a genetic algorithm according to an embodiment of the present invention comprises the following steps:

[0012] S1, initialize the target matrix and define the optical transformation of the MZI network;

[0013] S2, designing the MZI network layout based on the target matrix and generating a preliminary MZI network structure;

[0014] S3, generating an initial population, and randomly initializing the phase shifter parameters of each individual in the initial population;

[0015] S4, defining a fitness function, and calculating a fitness value according to the phase shifter parameter configuration of each individual in the initial population;

[0016] S5. According to the fitness value, the roulette wheel method is used to select individuals with high fitness;

[0017] S6, perform a crossover operation on the selected individuals, exchange the phase shifter parameters of the parent individuals to generate a new generation of individuals, and update the population;

[0018] S7, performing mutation operation, making small-range random adjustments to some phase shifter parameters of the individual, incorporating the mutated individual into the current population and updating the population;

[0019] S8. Repeat steps S4-S7 to gradually optimize the individuals in the population through multiple iterations, and determine whether the stopping condition is met according to the convergence condition. If the stopping condition is met, output the optimal solution.

[0020] Optionally, the target matrix is ​​an N*N complex matrix, describing the phase and amplitude relationship between the input optical signal and the output optical signal.

[0021] Optionally, the MZI network implements matrix transformation through a series of MZI arrays.

[0022] Optionally, the S3 specifically includes:

[0023] S31, determining an initial population size, wherein the selection of the initial population size is based on the complexity and optimization requirements of the MZI network to ensure that the population has high diversity and global search capability;

[0024] S32, generating each individual in the initial population, wherein the individual represents a set of phase shifter parameter configurations, and each individual corresponds to a parameter set controlled by multiple phase shifter voltages:

[0025] V=(v 1 ,v 2 ,...,v k );

[0026] Among them, V is each individual, v i is the voltage value of phase shifter i, k is the number of phase shifters in the MZI network, and the voltage value of the phase shifter is randomly generated by the formula:

[0027] v i = rand(v min ,v max );

[0028] Among them, v min is the minimum value of the phase shifter voltage, v max is the maximum value of the phase shifter voltage.

[0029] Optionally, the S4 specifically includes:

[0030] S41. Define the fitness function to quantify the degree to which each individual achieves the target matrix transformation of the MZI network:

[0031]

[0032] Among them, ∥OT∥ 2 It is the L2 norm square between the output matrix O and the target matrix T, and the calculation formula is:

[0033]

[0034] Among them, ij is the element in the output matrix O, t ij is the element in the target matrix T;

[0035] S42. According to each individual phase shifter parameter configuration V, the generated output matrix O is calculated:

[0036] O = MV;

[0037] Wherein, M is the optical transformation matrix of the MZI network;

[0038] S43. Calculate the fitness value F according to the difference between the output matrix O and the target matrix T. The larger the fitness value, the smaller the difference between the phase shifter parameter configuration V of the current individual and the target matrix T, that is, the current individual better achieves the target optical transformation.

[0039] Optionally, the S5 specifically includes:

[0040] S51. Define selection probability:

[0041]

[0042] Among them, P(i) is the selection probability, F i is the fitness value of individual i, is the sum of the fitness values of all individuals in the population, and N is the total number of individuals in the population;

[0043] S52. Select individuals according to the selection probability P(i) and select a certain number of individuals with higher fitness.

[0044] Optionally, the S6 specifically includes:

[0045] S61. Select the parental individuals to be crossed from the individuals output in the S5 step;

[0046] S62. Select the crossover method. The crossover method adopts single-point crossover or multi-point crossover. The crossover point k of the single-point crossover is generated by a random number, and the range is 1 ≤ k < N. The multi-point crossover selects multiple crossover points for multiple exchanges to further increase the diversity of the population;

[0047] S63. Determine the crossover rate, and the crossover rate is defined as the proportion of the parental pairs selected for the crossover operation;

[0048] S64. Determine the position of the crossover point. Taking this position as the segmentation point, exchange the parameters of parental individual 1 and parental individual 2 to generate new offspring individuals:

[0049]

[0050] Among them, are the first k phase shifter parameters of parental individual 1, are the last N - k phase shifter parameters of parental individual 1, are the first k phase shifter parameters of parental individual 2, are the last N - k phase shifter parameters of parental individual 2;

[0051] S65. Update the population.

[0052] Optionally, the S7 specifically includes:

[0053] S71. Select the individuals to be mutated from the individuals output in the S5 step;

[0054] S72. Set the execution probability of the mutation operation;

[0055] S73. According to the execution probability of the mutation operation, randomly select a certain phase shifter parameter v i in the individual for mutation and perform random adjustment:

[0056] v′ i = v i + Δv i ;

[0057] Among them, Δv iis a random disturbance within a preset range, obeying a Gaussian distribution, v′ i The phase shifter parameters generated after mutation;

[0058] S74. Incorporate the mutated individuals into the current population and update the population.

[0059] Optionally, the S8 specifically includes:

[0060] S81. According to the fitness function, the fitness of the new population is calculated, and selection, crossover and mutation operations are performed to gradually optimize the individuals in the population through multiple iterations;

[0061] S82, determining whether the number of iterations reaches the set maximum number of iterations, if so, stopping the algorithm immediately;

[0062] S83, judging whether the value of the fitness function reaches a set threshold;

[0063] S84. If the value of the fitness function reaches the set threshold or the number of iterations reaches the maximum number of iterations, the optimal solution is output.

[0064] According to an embodiment of the present invention, a device for self-optimization of an MZI matrix based on a genetic algorithm includes:

[0065] Initialization module, used to initialize the target matrix, design the MZI network layout, and generate the preliminary MZI network structure;

[0066] A population generation module, used to generate an initial population;

[0067] A fitness calculation module, used for calculating a fitness value according to the phase shifter parameter configuration of each individual in the initial population;

[0068] Individual selection, crossover, and mutation modules are used to perform selection, crossover, and mutation operations on individuals in the population;

[0069] The iterative control module is used to determine whether the stopping condition is met according to the convergence condition, and output the optimal solution if the stopping condition is met.

[0070] The beneficial effects of the present invention are:

[0071] First, the use of genetic algorithms as optimization tools can search for the optimal solution in a global scope, avoiding the defect of traditional local optimization methods that are prone to fall into the local optimal solution. This method not only enhances the diversity of the search process by simulating the selection, crossover and mutation mechanisms in the natural evolution process, but also accelerates the convergence speed of the optimization process, thereby greatly improving the optimization efficiency of the MZI network.

[0072] Secondly, the present invention quantifies the degree of realization of the target matrix transformation by each individual by defining a fitness function, ensuring that the optimization process can more accurately reflect the influence of each phase shifter parameter configuration in the network on the optical transformation. By introducing the dynamic adjustment fitness function and multi-scale optimization mechanism, the global search capability and convergence of the algorithm are further improved. Especially when dealing with complex optical transformation tasks, it can effectively deal with the interaction and optical coherence changes between the phase shifters in the MZI network, thereby improving the stability and accuracy of the optimization.

[0073] The feedback mechanism of fitness value is introduced to realize the precise control of the optimization process. When the optimization reaches the predetermined threshold, the algorithm can stop automatically, avoiding unnecessary calculations and improving the real-time and intelligent level of the system. This feature makes the present invention not only advantageous in theory, but also more operable and valuable in practical applications.

[0074] In summary, the genetic algorithm-based MZI network matrix self-optimization method provided by the present invention can effectively improve the optimization effect and stability of the MZI network in complex optical transformations, optimize the calculation accuracy, convergence speed and calculation efficiency, so that the MZI network can better adapt to the needs of high-dimensional matrix operations and complex system environments, and provide more efficient and accurate technical support for applications in photonic computing and related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] 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:

[0076] Figure 1 A flowchart of a method for self-optimization of MZI matrix based on genetic algorithm proposed by the present invention;

[0077] Figure 2 A schematic diagram of the crossover operation of a method for self-optimization of MZI matrix based on a genetic algorithm proposed by the present invention. DETAILED DESCRIPTION

[0078] 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.

[0079] refer to Figure 1 and Figure 2 , a method for self-optimization of MZI matrix based on genetic algorithm, comprising the following steps:

[0080] S1, initialize the target matrix and define the optical transformation of the MZI network;

[0081] S2, designing the MZI network layout based on the target matrix and generating a preliminary MZI network structure;

[0082] S3, generating an initial population, and randomly initializing the phase shifter parameters of each individual in the initial population;

[0083] S4, defining a fitness function, and calculating a fitness value according to the phase shifter parameter configuration of each individual in the initial population;

[0084] S5. According to the fitness value, the roulette wheel method is used to select individuals with high fitness;

[0085] S6, perform a crossover operation on the selected individuals, exchange the phase shifter parameters of the parent individuals to generate a new generation of individuals, and update the population;

[0086] S7, performing mutation operation, making small-range random adjustments to some phase shifter parameters of the individual, incorporating the mutated individual into the current population and updating the population;

[0087] S8. Repeat steps S4-S7 to gradually optimize the individuals in the population through multiple iterations, and determine whether the stopping condition is met according to the convergence condition. If the stopping condition is met, output the optimal solution.

[0088] In this implementation, the target matrix is ​​an N*N complex matrix, which describes the phase and amplitude relationship between the input optical signal and the output optical signal.

[0089] In this implementation, the MZI network implements matrix transformation through a series of MZI arrays.

[0090] In this implementation, S3 specifically includes:

[0091] S31, determining an initial population size, wherein the selection of the initial population size is based on the complexity and optimization requirements of the MZI network to ensure that the population has high diversity and global search capability;

[0092] S32, generating each individual in the initial population, wherein the individual represents a set of phase shifter parameter configurations, and each individual corresponds to a parameter set controlled by multiple phase shifter voltages:

[0093] V=(v 1 ,v 2 ,...,v k );

[0094] Among them, V is each individual, v i is the voltage value of phase shifter i, k is the number of phase shifters in the MZI network, and the voltage value of the phase shifter is randomly generated by the formula:

[0095] v i= rand(v min ,v max );

[0096] Among them, v min is the minimum value of the phase shifter voltage, v max is the maximum value of the phase shifter voltage.

[0097] In this implementation manner, the S4 specifically includes:

[0098] S41. Define the fitness function to quantify the degree to which each individual achieves the target matrix transformation of the MZI network:

[0099]

[0100] Among them, ∥OT∥ 2 It is the L2 norm square between the output matrix O and the target matrix T, and the calculation formula is:

[0101]

[0102] Among them, ij is the element in the output matrix O, t ij is the element in the target matrix T;

[0103] S42. According to each individual phase shifter parameter configuration V, the generated output matrix O is calculated:

[0104] O = MV;

[0105] Wherein, M is the optical transformation matrix of the MZI network;

[0106] S43. Calculate the fitness value F according to the difference between the output matrix O and the target matrix T. The larger the fitness value, the smaller the difference between the phase shifter parameter configuration V of the current individual and the target matrix T, that is, the current individual better achieves the target optical transformation.

[0107] In this implementation manner, S5 specifically includes:

[0108] S51. Define selection probability:

[0109]

[0110] Among them, P(i) is the selection probability, F i is the fitness value of individual i, is the sum of the fitness values ​​of all individuals in the population, and N is the total number of individuals in the population;

[0111] S52. Perform individual selection based on the selection probability P(i) to select a certain number of individuals with higher fitness.

[0112] In this embodiment, step S6 specifically includes:

[0113] S61. Select the parental individuals to be crossed from the individuals output in step S5;

[0114] S62. Select the crossing method. The crossing method adopts single-point crossing or multi-point crossing. The crossing point k of the single-point crossing is generated by a random number, and the range is 1 ≤ k < N. The multi-point crossing selects multiple crossing points for multiple exchanges to further increase the diversity of the population;

[0115] S63. Determine the crossing rate, which is defined as the proportion of parental pairs selected for the crossing operation;

[0116] S64. Determine the position of the crossing point. Taking this position as the dividing point, exchange the parameters of parental individual 1 and parental individual 2 to generate new offspring individuals:

[0117]

[0118] Wherein, are the first k phase shifter parameters of parental individual 1, are the last N - k phase shifter parameters of parental individual 1, are the first k phase shifter parameters of parental individual 2, are the last N - k phase shifter parameters of parental individual 2;

[0119] S65. Update the population.

[0120] In this embodiment, step S7 specifically includes:

[0121] S71. Select the individuals to be mutated from the individuals output in step S5;

[0122] S72. Set the execution probability of the mutation operation;

[0123] S73. According to the execution probability of the mutation operation, randomly select a certain phase shifter parameter v in the individual i for mutation and perform random adjustment:

[0124] v′ i = v i + Δv i ;

[0125] Wherein, Δv i is a random perturbation amount within a preset range, subject to a Gaussian distribution, and v′ i is the phase shifter parameter generated after mutation;

[0126] S74. Incorporate the mutated individuals into the current population and update the population.

[0127] In this implementation manner, S8 specifically includes:

[0128] S81. According to the fitness function, the fitness of the new population is calculated, and selection, crossover and mutation operations are performed to gradually optimize the individuals in the population through multiple iterations;

[0129] S82, determining whether the number of iterations reaches the set maximum number of iterations, if so, stopping the algorithm immediately;

[0130] S83, judging whether the value of the fitness function reaches a set threshold;

[0131] S84. If the value of the fitness function reaches the set threshold or the number of iterations reaches the maximum number of iterations, the optimal solution is output.

[0132] A device for self-optimization of MZI matrix based on genetic algorithm, comprising:

[0133] Initialization module, used to initialize the target matrix, design the MZI network layout, and generate the preliminary MZI network structure;

[0134] A population generation module, used to generate an initial population;

[0135] A fitness calculation module, used for calculating a fitness value according to the phase shifter parameter configuration of each individual in the initial population;

[0136] Individual selection, crossover, and mutation modules are used to perform selection, crossover, and mutation operations on individuals in the population;

[0137] The iterative control module is used to determine whether the stopping condition is met according to the convergence condition, and output the optimal solution if the stopping condition is met.

[0138] Embodiment 1:

[0139] This embodiment selects a typical optical communication system as an experimental scenario, aiming to demonstrate the application of the genetic algorithm-based MZI matrix self-optimization method in improving the performance of the optical communication system. In this system, the MZI network is used to realize the phase modulation and matrix transformation of the input optical signal. The goal of the experiment is to optimize the phase shifter parameters in the MZI network, increase the optical power, reduce noise, reduce the bit error rate, and improve the coherence of the optical signal, thereby enhancing the stability and transmission quality of the entire optical communication system. The experimental environment is set as an actual optical communication transmission link, with a network module composed of multiple MZI arrays, and the operating frequency band is 1550nm. In this scenario, the transmission quality of the optical signal directly affects the communication efficiency and stability. During the optimization process, we use genetic algorithms to automatically adjust the phase shifter parameters of the MZI network to maximize the system performance.

[0140] During the experiment, the optical power, noise level, bit error rate and coherence of the system were measured in detail, and the experiment was compared in two scenarios: before and after optimization.

[0141] Table 1 Performance comparison before and after optimization

[0142] Scenario Optical Power Noise Level Bit Error Rate Coherence Area A (before optimization) -22.5dBm 18.6dB <![CDATA[1.23×10 -3 ]]> 0.75 Area A (after optimization) -18.3dBm 15.2dB <![CDATA[3.56×10 -6 ]]> 0.92 Area B (before optimization) -21.8dBm 17.2dB <![CDATA[1.12×10 -3 ]]> 0.76 Area B (after optimization) -17.9dBm 14.9dB <![CDATA[2.98×10 -6 ]]> 0.91 Area C (before optimization) -23.1dBm 19.1dB <![CDATA[1.43×10 -3 ]]> 0.74 Area C (after optimization) -18.9dBm 16.5dB <![CDATA[3.12×10 -6 ]]> 0.93

[0143] First, from the perspective of the change in optical power, in all test areas, the optical power after optimization has increased compared to before optimization. In the test of area A, the optical power before optimization was -22.5dBm, and after optimization it increased to -18.3dBm, an increase of 4.2dB; in areas B and C, the optical power increased by 3.9dB and 4.2dB respectively. This change shows that the MZI network optimized by the genetic algorithm can significantly enhance the transmission power of the optical signal, improve the transmission efficiency of the system, reduce signal attenuation, and thus optimize the quality of optical communication.

[0144] Secondly, the noise level is generally reduced in the optimized results. In area A, the noise level was 18.6dB before optimization and dropped to 15.2dB after optimization, a decrease of 3.4dB; the noise levels in areas B and C decreased by 2.3dB and 2.6dB respectively. This shows that the optimized MZI network effectively suppresses the noise in the system and enhances the signal-to-noise ratio, thereby improving the stability and anti-interference ability of the system.

[0145] Let’s look at the change in bit error rate. In region A, the BER before optimization is 1.23×10 -3 , and after optimization it dropped to 3.56×10 -6 ; The bit error rate is also greatly improved in areas B and C. This significant decrease in bit error rate reflects that the optimized MZI network can provide more stable and accurate signal transmission, reduce the probability of errors in the transmission process, and further improve the reliability of the system.

[0146] Finally, the improvement in coherence is also very obvious. Before optimization, the coherence of region A was 0.75, and after optimization it increased to 0.92, an increase of 0.17; the coherence of regions B and C increased by 0.15 and 0.19 respectively. This means that the MZI network optimized by the genetic algorithm performs better in phase modulation and optical signal synchronization, and the coherence of the system is enhanced, thereby optimizing the stability and transmission quality of the signal.

[0147] In summary, it can be seen from the experimental data that the MZI matrix self-optimization method based on genetic algorithm has significant advantages in improving optical power, reducing noise, reducing bit error rate and improving coherence. The optimized system has shown significant improvements over the pre-optimization system in various key indicators, further proving the effectiveness and feasibility of the method of the present invention in improving the performance of optical communication systems.

[0148] 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 method for self-optimization of MZI matrix based on genetic algorithm, characterized in that: It includes the following steps: S1. Initialize the target matrix and define the optical transformation of the MZI network; S2. Design the MZI network layout based on the target matrix to generate a preliminary MZI network structure; S3. Generate an initial population and randomly initialize the phase shifter parameters for each individual in the initial population; S4. Define a fitness function and calculate the fitness value according to the phase shifter parameter configuration of each individual in the initial population; S5. Select individuals with high fitness using the roulette wheel method according to the fitness value; S6. Perform a crossover operation on the selected individuals, exchange the phase shifter parameters of the parent individuals to generate new individuals of the new generation, and update the population; S7. Perform a mutation operation, randomly adjust the phase shifter parameters of some individuals within a small range, incorporate the mutated individuals into the current population, and update the population; S8. Repeat steps S4 - S7, gradually optimize the individuals in the population through multiple iterations, and judge whether the stopping condition is met according to the convergence condition. If the stopping condition is met, output the optimal solution.

2. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The target matrix is an N*N complex matrix that describes the phase and amplitude relationship between the input optical signal and the output optical signal.

3. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The MZI network realizes matrix transformation through a series of MZI arrays.

4. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The specific content of S3 includes: S31. Determine the size of the initial population. The selection of the size of the initial population is based on the complexity and optimization requirements of the MZI network to ensure that the population has high diversity and global search ability; S32. Generate each individual in the initial population. The individual represents a set of phase shifter parameter configurations, and each individual corresponds to a parameter set controlled by multiple phase shifter voltages: V=(v1,v2,...,v k ); Among them, V is each individual, v i is the voltage value of phase shifter i, k is the number of phase shifters in the MZI network, and the voltage value of the phase shifter is randomly generated by the formula: v i =rand(v min ,v max ); Among them, v min is the minimum value of the phase shifter voltage, v max is the maximum value of the phase shifter voltage.

5. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The specific content of S4 includes: S41. Define a fitness function to quantify the degree of realization of the target matrix transformation of the MZI network for each individual: Among them, ∥OT∥ 2 It is the L2 norm square between the output matrix O and the target matrix T, and the calculation formula is: Among them, ij is the element in the output matrix O, t ij is the element in the target matrix T; S42. Calculate the generated output matrix O according to the phase shifter parameter configuration V of each individual: O = MV; where M is the optical transformation matrix of the MZI network; S43. Calculate the fitness value F according to the difference between the output matrix O and the target matrix T. The larger the fitness value, the smaller the difference between the phase shifter parameter configuration V of the current individual and the target matrix T, that is, the current individual better realizes the target optical transformation.

6. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The specific content of S5 includes: S51. Define the selection probability: Among them, P(i) is the selection probability, F i is the fitness value of individual i, is the sum of the fitness values ​​of all individuals in the population, and N is the total number of individuals in the population; S52. Select individuals according to the selection probability P(i) and select a certain number of individuals with higher fitness.

7. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The specific content of S6 includes: S61. Select the parent individuals to be crossed from the individuals output in step S5; S62. Select the crossover method. The crossover method adopts single-point crossover or multi-point crossover. The crossover point k of the single-point crossover is generated by a random number, and the range is 1 ≤ k < N. The multi-point crossover selects multiple crossover points for multiple exchanges to further increase the diversity of the population; S63. Determine the crossover rate. The crossover rate is defined as the proportion of parent pairs selected for the crossover operation; S64. Determine the position of the crossover point. Taking this position as the splitting point, exchange the parameters of parent individual 1 and parent individual 2 to generate new offspring individuals: in, are the first k phase shifter parameters of parent individual 1, are the parameters of the last Nk phase shifters of parent individual 1, are the first k phase shifter parameters of parent individual 2, are the parameters of the last Nk phase shifters of parent individual 2; S65. Update the population.

8. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The specific content of S7 includes: S71, selecting individuals to be mutated from the individuals output in step S5; S72, setting the execution probability of the mutation operation; S73. According to the execution probability of the mutation operation, a phase shifter parameter v in the individual is randomly selected. i Perform mutation and make random adjustments: v′ i =v i +Δv i ; Where Δv i is a random disturbance within a preset range, obeying a Gaussian distribution, v′ i The phase shifter parameters generated after mutation; S74. Incorporate the mutated individuals into the current population and update the population.

9. The method for self-optimization of MZI matrix based on genetic algorithm according to claim 1, characterized in that: The S8 specifically includes: S81. According to the fitness function, the fitness of the new population is calculated, and selection, crossover and mutation operations are performed to gradually optimize the individuals in the population through multiple iterations; S82, determining whether the number of iterations reaches the set maximum number of iterations, if so, stopping the algorithm immediately; S83, judging whether the value of the fitness function reaches a set threshold; S84. If the value of the fitness function reaches the set threshold or the number of iterations reaches the maximum number of iterations, the optimal solution is output.

10. A device for self-optimization of MZI matrix based on genetic algorithm, executing the method for self-optimization of MZI matrix based on genetic algorithm according to any one of claims 1 to 9, characterized in that: include: Initialization module, used to initialize the target matrix, design the MZI network layout, and generate the preliminary MZI network structure; A population generation module, used to generate an initial population; A fitness calculation module, used for calculating a fitness value according to the phase shifter parameter configuration of each individual in the initial population; Individual selection, crossover, and mutation modules are used to perform selection, crossover, and mutation operations on individuals in the population; The iterative control module is used to determine whether the stopping condition is met according to the convergence condition, and output the optimal solution if the stopping condition is met.

Images