Optimization method for reverse design of photonic devices

Abstract

The present application relates to the technical field of photonic device design, and especially relates to a photonic device reverse design optimization method, the present application adopts sin mapping, segmented linear chaotic mapping and reverse learning combined mode to generate initial population, enhances initial population diversity, reduces initial error and accelerates algorithm convergence; adopts Levy flight and teaching mechanism alternating parallel mode to accelerate algorithm convergence process, and adopts teaching factor decreasing with iteration number to further accelerate algorithm to find optimal solution; designs dynamic self-adaptive discovery probability, discovery probability linearly decreases in search late stage, and local optimal solution is more easily discovered, thereby accelerating local convergence; differential evolution retains individuals with higher fitness, so that population continuously approaches optimal solution, thereby increasing algorithm convergence speed; in addition, the present application adopts differential evolution mechanism to generate new individuals through differential calculation on individuals in population, thereby increasing population diversity and global search capability.

Description

Technical Field

[0001] The present invention belongs to the technical field of photonic device design, and in particular relates to a photonic device reverse design optimization method. Background Art

[0002] With the rapid development of information technology and photonics, photonic devices have been widely used in fields such as communications, sensing, computing, and imaging. In particular, in the field of silicon photonics, silicon-based photonic devices have become the core of photonic integrated circuit (PIC) design due to their low cost, high performance, and compatibility with existing integrated circuit technologies. However, as silicon photonic device design becomes increasingly complex, traditional design methods (such as empirical forward design methods) are no longer able to meet the requirements for efficient and high-precision photonic device design.

[0003] Traditional photonic device design methods typically rely on the designer's experience and understanding of physical principles to improve device performance through iterative optimization. While effective to a certain extent, this approach faces significant limitations, including the complexity of the design space, significant computational resource consumption, and the potential for optimization results to be confined to the designer's knowledge boundaries. Furthermore, traditional design methods are unable to adapt to the diverse demands of device structure and functionality, particularly in high-dimensional parameter spaces, where the efficiency and quality of optimization are often limited.

[0004] Optimization algorithms play a crucial role in the reverse design of photonic devices. They optimize device performance by optimizing design parameters. Photonic device design based on optimization algorithms offers the advantages of efficiently searching the design space, avoiding local optimal solutions, accelerating convergence, and improving design accuracy. However, existing algorithms have the following shortcomings:

[0005] (1) Slow optimization speed: Traditional optimization algorithms, such as genetic algorithm (GA) and particle swarm optimization (PSO), converge slowly in high-dimensional design spaces. Especially when a large number of design parameters need to be processed in photonic device design, the optimization process consumes a lot of time.

[0006] (2) Easy to fall into local optimal solutions: Whether it is the GA algorithm or the PSO optimization algorithm, when facing complex photonic device design problems, due to its randomness, it often falls into local optimal solutions, resulting in the final design failing to achieve optimal performance. This local optimal phenomenon is more significant in high-dimensional design spaces (designs with a large number of parameters). Summary of the Invention

[0007] In light of this, the present invention aims to provide a reverse design optimization method for photonic devices. This method uses an optimization algorithm to reversely derive a suitable design structure from a given device's target performance. Compared to traditional forward design, this method allows for greater design freedom in exploring optimization solutions, leading to more innovative and efficient photonic device designs.

[0008] To achieve the above object, the technical solution created by the present invention is implemented as follows:

[0009] A photonic device reverse design optimization method, comprising:

[0010] S1: Generate the -1 generation device parameter vector of the photonic device to be designed by using sin mapping and piecewise linear chaotic mapping; then perform reverse learning on the -1 generation device parameter vector to generate the initial generation device parameter vector of the photonic device; obtain the initial parameter population of the photonic device based on the initial generation device parameter vector; and set the number of optimization iterations.

[0011] S2: Based on the number of optimization iterations set in step S1, the next generation of candidate individuals is found from the initial parameter population obtained in step S1 by alternating Levy flight and teaching mechanisms in parallel.

[0012] S3: Based on the number of optimization iterations set in step S1, adaptive discovery is performed on all next-generation candidate individuals obtained in step S2 to obtain a new generation of random population that meets the adaptive discovery probability;

[0013] S4: Perform differential evolution on the individuals in the new generation random population obtained in step S3 one by one to obtain an evolved population, and record the optimal solution in the evolved population;

[0014] S5: Replace the initial parameter population in step S2 with the evolved population obtained in step S4, and repeat steps S2 to S4 for optimization iterations with the number of optimization iterations set in step S1 to obtain all optimal solutions.

[0015] Furthermore, step S1 includes:

[0016] S11: randomly generate a -2 generation device parameter vector, where the dimension of the -2 generation device parameter vector is the same as the number of device parameter vectors of the photonic device;

[0017] S12: Obtain a -1 generation device parameter vector based on the -2 generation device parameter vector obtained in step S11 by using sin mapping and piecewise linear chaotic mapping;

[0018] S13: performing reverse learning on the -1 generation device parameter vector obtained in step S12 to generate a first generation device parameter vector of the photonic device;

[0019] S14: Set the number of individuals in the initial parameter population according to the photonic device, repeat steps S11 to S13, and obtain the initial device parameter vector in the initial parameter population.

[0020] Furthermore, in step S12, three first random numbers that meet the standard uniform distribution are randomly generated, and the -1 generation device parameter vector is obtained from the -2 generation device parameter vector using the following formula:

[0021] ;

[0022] Among them, r1,1, and Represent the three first random numbers respectively, represents the parameter vector of the -1 generation device, Represents the parameter vector of the -2 generation device, and mod represents the modular operation.

[0023] Furthermore, in step S13:

[0024] Setting the search interval upper limit, search interval lower limit and search interval midpoint according to the photonic device;

[0025] A second random number that satisfies the standard uniform distribution is randomly generated, and combined with the upper limit and midpoint of the search interval, the initial device parameter vector is obtained by the following formula:

[0026] ;

[0027] in, represents the initial device parameter vector, Indicates the upper limit of the search interval, represents the midpoint of the search interval, Represents the second random number.

[0028] Furthermore, step S2 includes:

[0029] Randomly generate a third random number that satisfies a standard uniform distribution;

[0030] If the third random number is less than the candidate threshold, the Levy flight principle is used to find the next generation of candidate individuals; if the third random number is greater than or equal to the candidate threshold, the teaching mechanism is used to find the next generation of candidate individuals.

[0031] Furthermore, the process of finding next generation candidate individuals using the Levy flight principle includes:

[0032] The next generation of candidate individuals is obtained by the following formula:

[0033] ;

[0034] in, represents the i-th candidate individual in the t+1-th optimization iteration, i={2,3,...n}, Indicates the number of individuals, represents the i-th candidate individual in the t-th optimization iteration, represents the step size scaling factor, represents the Levy random flight path, represents the flight control factor, represents point-to-point multiplication;

[0035] Levy random flight paths It is obtained by the following formula:

[0036] ;

[0037] in, and represents a random vector that satisfies the standard normal distribution, To calculate the Levy random flight path The intermediate variable is:

[0038] ;

[0039] in, Represents the Gamma function.

[0040] Furthermore, the process of using the teaching mechanism to find the next generation of candidate individuals includes:

[0041] Randomly generate a first random array that satisfies a uniform distribution in the interval [1,2]. The number of random numbers in the first random array is equal to the number of optimization iterations. The maximum value in the first random array is the upper limit of the teaching factor, and the minimum value in the first random array is the lower limit of the teaching factor.

[0042] The next generation of candidate individuals is obtained by the following formula:

[0043] ;

[0044] in, represents the teaching factor, represents the average value of all candidate individuals in the contemporary era, represents the candidate individual with the best objective function among all contemporary candidate individuals, and Represent the upper limit and lower limit of teaching factor respectively, represents the number of optimization iterations, and R represents a standard uniformly distributed random number.

[0045] Furthermore, the process of step S3 includes:

[0046] A second random number array satisfying a uniform distribution is randomly generated in the interval (0, 1), and the number of random numbers in the second random number array is equal to the number of optimization iterations;

[0047] The discovery probability benchmark is determined by the following formula:

[0048] ;

[0049] wherein, represents the adaptive discovery probability at the tth optimization iteration, and respectively represent the maximum value and the minimum value in the random number array;

[0050] The adaptive discovery probability of the ith candidate individual is determined by the following formula:

[0051] ;

[0052] wherein, represents the adaptive discovery probability of the ith candidate individual, is the fitness of the ith candidate individual, and respectively represent the maximum value and the minimum value of the fitness of all candidate individuals; the fitness is calculated by a target function;

[0053] For each candidate individual, a fourth random number satisfying a standard uniform distribution is randomly generated;

[0054] If the adaptive discovery probability of the current candidate individual is greater than or equal to the fourth random number, the current random individual is not generated; if the adaptive discovery probability of the current candidate individual is less than the fourth random number, the current random individual is generated by the following formula:

[0055] ;

[0056] wherein, represents the ith current random population individual, L represents the lower limit of the search interval, and d represents the number of device parameters of the photonic device, represents a dot product, and rand represents a function of generating a random number;

[0057] The target function of each current candidate individual and each current random individual is calculated, and the two target function values are compared correspondingly: if the current candidate individual is better, the current candidate individual is retained; otherwise, the corresponding candidate individual is replaced with the random individual to form a new generation of random population.

[0058] Further, step S4 comprises:

[0059] S41: sequentially selecting one base individual in the new generation of random population, and randomly selecting three individuals from the current random population;

[0060] S42: Based on the three individuals selected in step S41, generate mutant individuals using the following formula:

[0061] ;

[0062] in, represents the mutation individual at the t-th optimization iteration, 、 and represents the three individuals selected in step S41, and F represents the mutation probability;

[0063] S43: For each element in the base individual selected in step S41, generate a fifth random number that satisfies the standard uniform distribution; based on the fifth random number, replace the element in the base individual with the element in the variant individual obtained in step S42 to obtain a crossover individual;

[0064] S44: Compare the objective function value of the base individual with the crossover individual obtained in step S43: if the objective function value of the base individual is better than the objective function value of the crossover individual, retain the base individual; otherwise, replace the base individual with the crossover individual;

[0065] S45: Repeat steps S41 to S44 until all individuals in the new generation random population are replaced to obtain an evolved population.

[0066] Furthermore, in step S43, the crossover individual is obtained by the following formula:

[0067] ;

[0068] in, represents the jth element in the kth basic individual in the current random population, represents the jth element in the mutant individual at the tth optimization iteration, represents the jth element in the crossover individual at the tth optimization iteration, represents the fifth random number corresponding to each element in the basic individual, and C represents the crossover probability.

[0069] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0070] In the photonic device reverse design optimization method created by the present invention, the initial population is first generated by combining sin mapping, piecewise linear chaos mapping and reverse learning, thereby enhancing the diversity of the initial population, reducing the initial error and accelerating the convergence of the algorithm. In addition, the present invention adopts an advantageous upper teaching and learning mechanism with fewer parameter settings and faster solution speed, and combines it with Levy flight to accelerate the convergence process of the algorithm. The present invention designs a dynamic adaptive discovery probability, and the discovery probability shows a dynamic decreasing trend with the number of iterations, so that the discovery probability is larger in the early search stage, enhancing the possibility of global search, and linearly decreases in the later stage, making the local optimal solution easier to be discovered. Based on this, the present invention not only increases the global search capability but also accelerates the convergence of the algorithm. Finally, the present invention generates new individuals by performing differential calculations on individuals in the population, thereby increasing the diversity of the population and the global search capability. In addition, individuals with higher fitness are retained, so that the population continues to approach the optimal solution, thereby increasing the convergence speed of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0072] Figure 1 A schematic diagram of a process for the reverse design optimization method of a photonic device according to an embodiment of the present invention;

[0073] Figure 2 A schematic diagram of the process of step S1 according to an embodiment of the present invention;

[0074] Figure 3 This is a flow chart of step S2 according to an embodiment of the present invention;

[0075] Figure 4 This is a flow chart of step S3 according to an embodiment of the present invention;

[0076] Figure 5 This is a flow chart of step S4 according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0078] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0079] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second" and the like are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, features defined as "first", "second" and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0080] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art can understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0081] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0082] like Figures 1 to 5 As shown, the photonic device reverse design optimization method described in the embodiment of the present invention includes:

[0083] S1: Generate the -1 generation device parameter vector of the photonic device to be designed using a sin mapping and piecewise linear chaotic mapping method. Then, perform reverse learning on the -1 generation device parameter vector to generate the initial generation device parameter vector of the photonic device. Based on the initial generation device parameter vector, obtain the initial parameter population of the photonic device. Simultaneously, set the number of optimization iterations. A higher number of optimization iterations results in better optimization results, but also requires more computation. In this embodiment, this number is set to 20.

[0084] In some embodiments, step S1 includes:

[0085] S11: Randomly generate a -2 generation device parameter vector, where the dimension of the -2 generation device parameter vector is the same as the number of device parameter vectors of the photonic device.

[0086] In the embodiment of the present invention, the parameter vector of the -2 generation device is expressed as , where d represents the number of device parameters of the photonic device. For example, in the TE and TM mode beam splitters, the number of device parameters is d=20×20=400.

[0087] S12: Using sin mapping and piecewise linear chaotic mapping, the -1 generation device parameter vector is obtained from the -2 generation device parameter vector obtained in step S11. In step S12, three first random numbers that meet the standard uniform distribution are randomly generated, and the -1 generation device parameter vector is obtained from the -2 generation device parameter vector using the following formula:

[0088] ;

[0089] Among them, r1,1, and Represent the three first random numbers respectively, Indicates the -1 generation device parameter vector, mod indicates the modulus operation. In the embodiment of the present invention, mod(x,1) means dividing x by 1 and taking the remainder so that it remains within the range of [0,1]. For example, if the -2 generation device parameter vector Parameters in If it exceeds the interval [0,1], it can be mapped back to the interval through the mod operation to ensure that the parameter The value will not exceed the preset range [0,1].

[0090] S13: Perform reverse learning on the -1 generation device parameter vector obtained in step S12 to generate the initial generation device parameter vector of the photonic device. In step S13:

[0091] The search interval upper limit, search interval lower limit, and search interval midpoint are set based on the photonic device. In an embodiment of the present invention, the search interval upper limit U and search interval midpoint M are determined based on process limitations or performance requirements. For example, if the minimum size of microelectronic processing is 20 nm, and the size of each grid after gridding is 120 nm, i.e., the maximum size is 120 nm, then U is 120 nm and M is 0.5 × (120 nm + 20 nm) = 70 nm.

[0092] A second random number that satisfies the standard uniform distribution is randomly generated, and combined with the upper limit and midpoint of the search interval, the initial device parameter vector is obtained by the following formula:

[0093] ;

[0094] in, represents the initial device parameter vector, Represents the second random number.

[0095] S14: Set the number of individuals in the initial parameter population according to the photonic device, repeat steps S11 to S13, and obtain the initial device parameter vector in the initial parameter population. In the embodiment of the present invention, the dimension of the initial parameter population is d, and the population contains n individuals, that is, the number of individuals is n. It can be understood that the initial parameter population is expressed as:

[0096] ;

[0097] in, represents the initial parameter population, represents the nth generation device parameter vector, Represents the dth parameter in the nth generation device parameter vector.

[0098] S2: Based on the number of optimization iterations set in step S1, the next generation of candidate individuals is found from the initial parameter population obtained in step S1 by alternating Levy flight and teaching mechanisms in parallel.

[0099] In some embodiments, step S2 includes:

[0100] A third random number is randomly generated that satisfies a standard uniform distribution. If the third random number is less than the candidate threshold, the Levy flight principle is used to find the next generation of candidate individuals. If the third random number is greater than or equal to the candidate threshold, the teaching mechanism is used to find the next generation of candidate individuals. The candidate threshold is adaptively adjusted based on actual conditions; in this embodiment, it is set to 0.5.

[0101] In some embodiments, the process of finding next generation candidate individuals using the Levy flight principle includes:

[0102] The next generation of candidate individuals is obtained by the following formula:

[0103] ;

[0104] in, represents the i-th candidate individual in the t+1-th optimization iteration, i={2,3,...n}, represents the i-th candidate individual in the t-th optimization iteration, Represents the step size scaling factor, which is used to adjust the search range of the step size. represents the Levy random flight path, represents the flight control factor, represents point-to-point multiplication, The role of Levy random flight path Each element in the scale times. Step size scaling factor and flight control factors Adaptively adjust according to actual conditions. In this embodiment of the present invention, the step size scaling factor , flight control factor Levy random flight path It is obtained by the following formula:

[0105] ;

[0106] in, and represents a random vector that satisfies the standard normal distribution, the random vector and is a 1×d row vector, and each element in the random vector conforms to the normal distribution. To calculate the Levy random flight path The intermediate variable is:

[0107] ;

[0108] in, Represents the Gamma function.

[0109] In some embodiments, the process of finding next generation candidate individuals using the teaching mechanism includes:

[0110] Randomly generate a first random array that satisfies a uniform distribution in the interval [1,2]. The number of random numbers in the first random array is equal to the number of optimization iterations. The maximum value in the first random array is the upper limit of the teaching factor, and the minimum value in the first random array is the lower limit of the teaching factor.

[0111] The next generation of candidate individuals is obtained by the following formula:

[0112] ;

[0113] in, represents the teaching factor, represents the average value of all candidate individuals in the contemporary era, represents the candidate individual with the best objective function among all contemporary candidate individuals, and Represent the upper limit and lower limit of teaching factor respectively, Represents the number of optimization iterations, and R represents a standard uniformly distributed random number. The objective function is determined according to the photonic device to be designed. The objective function in the embodiment of the present invention is based on the TE and TM mode beam splitters. The process of calculating the objective function value is: simulating the device performance according to individual parameters and calculating the objective function value. For example, for TE and TM mode beam splitters, for TM output waveguides, the objective function is defined as TM power / TE power. The larger the objective function, the better the effect. Teaching factor Adaptive adjustment is performed in the form of linear decrease as the number of iterations t increases, further accelerating the algorithm to find the optimal solution. In the embodiment of the present invention, adaptive adjustment refers to: teaching factor It decreases in a linear decreasing manner as the number of iterations t increases.

[0114] S3: Based on the number of optimization iterations set in step S1, adaptive discovery is performed on all next-generation candidate individuals obtained in step S2 to obtain a new generation of random population that meets the adaptive discovery probability.

[0115] In some embodiments, the process of step S3 includes:

[0116] A second random number array that satisfies uniform distribution is randomly generated in the interval (0, 1), and the number of random numbers in the second random number array is equal to the number of optimization iterations. In the embodiment of the present invention, the range of the second random number array is between 0.005 and 0.995.

[0117] The probability of discovery benchmark is determined by the following formula:

[0118] ;

[0119] in, represents the discovery probability benchmark at the t-th optimization iteration, and Represents the maximum and minimum values ​​in the random array, the maximum value and minimum value These are the upper and lower limits of the discovery probability benchmark.

[0120] The adaptive discovery probability of the i-th candidate individual is determined by the following formula:

[0121]

[0122] in, The adaptive discovery probability of the i-th candidate individual, is the fitness of the i-th candidate individual, and are the maximum and minimum fitness values ​​of all candidate individuals respectively. The fitness is calculated by the objective function.

[0123] For each candidate individual, a fourth random number that satisfies the standard uniform distribution is randomly generated. If the adaptive discovery probability of the current candidate individual is greater than or equal to the fourth random number, the current random individual is not generated; if the adaptive discovery probability of the current candidate individual is less than the fourth random number, the current random individual is generated using the following formula:

[0124] ;

[0125] in, Represents the current random individual, L represents the lower limit of the search interval, represents dot product, and rand represents a function for generating random numbers.

[0126] Calculate each current candidate individual and the current random individual The objective function is calculated and the two objective function values ​​are compared accordingly: if the current individual is better, the current candidate individual is retained, otherwise the corresponding candidate individual is replaced by a random individual to form a new generation of random population.

[0127] S4: Perform differential evolution on the individuals in the new generation random population obtained in step S3 one by one to obtain an evolved population, and record the optimal solution in the evolved population.

[0128] In some embodiments, step S4 includes:

[0129] S41: Sequentially select a base individual from the new generation random population and randomly select three individuals from the new generation random population.

[0130] S42: Based on the three individuals selected in step S41, generate mutant individuals using the following formula:

[0131] ;

[0132] in, represents the mutation individual at the t-th optimization iteration, 、 and represents the three individuals selected in step S41, and F represents the mutation probability, which is usually 0.5.

[0133] S43: For each element in the base individual selected in step S41, generate a fifth random number that satisfies the standard uniform distribution; based on the fifth random number, replace the element in the variant individual obtained in step S42 with the element in the base individual to obtain a crossover individual. The crossover individual is obtained by the following formula:

[0134] ;

[0135] in, represents the jth element in the kth basic individual in the new generation of random population, represents the jth element in the mutant individual at the tth optimization iteration, represents the jth element in the crossover individual at the tth optimization iteration, represents the fifth random number corresponding to each element in the basic individual, and C represents the crossover probability. In the embodiment of the present invention, the crossover probability C is 0.7.

[0136] S44: compare the objective function value of the basic individual with that of the crossed individual obtained in step S43; if the objective function value of the basic individual is better than that of the crossed individual, the basic individual is kept, otherwise the crossed individual replaces the basic individual.

[0137] S45: repeat steps S41-S44 until the replacement of all individuals in the new generation random population is completed, and an evolved population is obtained.

[0138] S5: replace the initial parameter population in step S2 with the evolved population obtained in step S4, and repeat steps S2-S4 for optimization iteration with the optimization iteration number set in step S1, and obtain all optimal solutions.

[0139] It should be understood that the various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present disclosure can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions of the present disclosure can be achieved, which is not limited herein.

[0140] The above detailed description does not constitute a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for reverse design optimization of photonic devices, characterized in that: include: S1: Randomly generate a -2 generation device parameter vector, and use sin mapping and piecewise linear chaotic mapping to generate a -1 generation device parameter vector of the photonic device to be designed based on the -2 generation device parameter vector; then reversely learn the -1 generation device parameter vector to generate a first generation device parameter vector of the photonic device; obtain an initial parameter population of the photonic device based on the first generation device parameter vector; and set the number of optimization iterations. S2: Based on the number of optimization iterations set in step S1, the next generation of candidate individuals is found from the initial parameter population obtained in step S1 by alternating Levy flight and teaching mechanisms in parallel. S3: Based on the number of optimization iterations set in step S1, adaptive discovery is performed on all next-generation candidate individuals obtained in step S2 to obtain a new generation of random population that meets the adaptive discovery probability; S4: performing differential evolution on the individuals in the new generation random population obtained in step S3 one by one to obtain an evolved population, and recording the optimal solution in the evolved population; S5: Replace the initial parameter population in step S2 with the evolved population obtained in step S4, and repeat steps S2 to S4 for optimization iterations with the number of optimization iterations set in step S1 to obtain all optimal solutions.

2. The method for reverse design optimization of photonic devices according to claim 1, characterized in that: Step S1 includes: S11: randomly generating a -2 generation device parameter vector, wherein the dimension of the -2 generation device parameter vector is the same as the number of device parameters of the photonic device; S12: Obtain the -1 generation device parameter vector according to the -2 generation device parameter vector obtained in step S11 by using sin mapping and piecewise linear chaotic mapping; S13: performing reverse learning on the -1 generation device parameter vector obtained in step S12 to generate a first generation device parameter vector of the photonic device; S14: setting the number of individuals in the initial parameter population according to the photonic device, and repeating steps S11 to S13 to obtain a first-generation device parameter vector in the initial parameter population.

3. The method for reverse design optimization of photonic devices according to claim 2, characterized in that: In step S12, three first random numbers that meet the standard uniform distribution are randomly generated, and the -1 generation device parameter vector is obtained according to the -2 generation device parameter vector using the following formula: ； Among them, r1,1, and Represent the three first random numbers respectively, represents the parameter vector of the -1 generation device, represents the parameter vector of the -2 generation device, and mod represents the modular operation.

4. The method for reverse design optimization of photonic devices according to claim 3, characterized in that: In step S13: Setting a search interval upper limit, a search interval lower limit, and a search interval midpoint according to the photonic device; A second random number that satisfies a standard uniform distribution is randomly generated, and the initial device parameter vector is obtained by combining the upper limit of the search interval and the midpoint of the search interval using the following formula: ； in, represents the initial device parameter vector, represents the upper limit of the search interval, represents the midpoint of the search interval, represents the second random number.

5. The method for reverse design optimization of photonic devices according to claim 4, characterized in that: Step S2 includes: Randomly generate a third random number that satisfies a standard uniform distribution; If the third random number is less than the candidate threshold, the next generation candidate individual is found using the Levy flight principle; if the third random number is greater than or equal to the candidate threshold, the next generation candidate individual is found using the teaching mechanism.

6. The method for reverse design optimization of photonic devices according to claim 5, characterized in that: The process of searching for the next generation candidate individuals using the Levy flight principle includes: The next generation candidate individuals are obtained by the following formula: ； in, represents the i-th candidate individual in the t+1-th optimization iteration, i={2,3,...n}, represents the number of individuals, represents the i-th candidate individual in the t-th optimization iteration, represents the step size scaling factor, represents the Levy random flight path, represents the flight control factor, represents point-to-point multiplication; Levy random flight paths It is obtained by the following formula: ； in, and represents a random vector that satisfies the standard normal distribution, To calculate the Levy random flight path The intermediate variable is: ； in, Represents the Gamma function.

7. The method for reverse design optimization of photonic devices according to claim 6, characterized in that: The process of finding the next generation candidate individuals using the teaching mechanism includes: Randomly generate a first random array that satisfies a uniform distribution in the interval [1, 2], wherein the number of random numbers in the first random array is equal to the number of optimization iterations, the maximum value in the first random array is the upper limit of the teaching factor, and the minimum value in the first random array is the lower limit of the teaching factor; The next generation candidate individuals are obtained by the following formula: ； in, represents the teaching factor, represents the average value of all candidate individuals in the contemporary era, represents the candidate individual with the best objective function among all contemporary candidate individuals, and Respectively represent the upper limit and lower limit of the teaching factor, represents the number of optimization iterations, and R represents a standard uniformly distributed random number.

8. The method for reverse design optimization of photonic devices according to claim 7, characterized in that: The process of step S3 includes: Randomly generate a second random array that satisfies a uniform distribution in the interval (0, 1), wherein the number of random numbers in the second random array is equal to the number of optimization iterations; The probability of discovery benchmark is determined by the following formula: ； in, represents the adaptive discovery probability at the t-th optimization iteration, and Respectively represent the maximum and minimum values ​​in the random array; The adaptive discovery probability of the i-th candidate individual is determined by the following formula: ； in, represents the adaptive discovery probability of the i-th candidate individual, is the fitness of the i-th candidate individual, and are the maximum and minimum fitness values ​​of all candidate individuals respectively; the fitness is calculated by the objective function; For each candidate individual, randomly generate a fourth random number that satisfies the standard uniform distribution; If the adaptive discovery probability of the current candidate individual is greater than or equal to the fourth random number, the current random individual is not generated; if the adaptive discovery probability of the current candidate individual is less than the fourth random number, the current random individual is generated by the following formula: ； in, represents the i-th individual in the current random population, L represents the lower limit of the search interval, d represents the number of device parameters of the photonic device, represents dot product, rand represents the function of generating random numbers; Calculate the objective function of each current candidate individual and each current random individual, and compare the two objective function values ​​accordingly: if the current candidate individual is better, retain the current candidate individual, otherwise replace the corresponding candidate individual with a random individual to form a new generation of random population.

9. The method for reverse design optimization of photonic devices according to claim 1, wherein: Step S4 includes: S41: Sequentially select a base individual from the next generation random population and randomly select three individuals from the current random population; S42: Based on the three individuals selected in step S41, generate mutant individuals using the following formula: ； in, represents the mutation individual at the t-th optimization iteration, 、 and represents the three individuals selected in step S41, and F represents the mutation probability; S43: For each element in the base individual selected in step S41, generate a fifth random number that satisfies the standard uniform distribution; based on the fifth random number, replace the element in the base individual with the element in the variant individual obtained in step S42 to obtain a crossover individual; S44: Compare the objective function value of the basic individual with the objective function value of the crossover individual obtained in step S43: if the objective function value of the basic individual is better than the objective function value of the crossover individual, retain the basic individual; otherwise, replace the basic individual with the crossover individual; S45: Repeat steps S41 to S44 until all individuals in the new generation random population are replaced to obtain the evolved population.

10. The method for reverse design optimization of photonic devices according to claim 9, characterized in that: In step S43, the crossover individual is obtained by the following formula: ； in, represents the jth element in the kth basic individual in the new generation of random population, represents the jth element in the mutant individual at the tth optimization iteration, represents the jth element in the crossover individual at the tth optimization iteration, represents the fifth random number corresponding to each element in the basic individual, and C represents the crossover probability.

Images