Glass composition optimization method and system based on neural network and goat optimization algorithm

Abstract

The invention discloses a glass composition optimization method and system based on a neural network and a goat optimization algorithm, and relates to the technical field of glass composition optimization. The method comprises the following steps: optimizing hyper-parameters of a neural network by using a goat optimization algorithm, and constructing a high-performance neural network prediction model; then, the optimized neural network prediction model is embedded into a fitness evaluation link of a goat optimization algorithm, in the glass component optimization process, performance indexes of different component combinations are rapidly predicted through a neural network, the performance indexes serve as fitness values of the goat optimization algorithm, and the algorithm is guided to efficiently search for the optimal component combination. According to the method, through the high-precision prediction capability of the neural network and the global search and adaptive capability of the goat optimization algorithm, the efficiency and accuracy of glass component design are remarkably improved, an innovative solution is provided for research and development of high-performance glass materials, the separation dilemma of'prediction-design 'of glass components is broken through, and the method is suitable for large-scale popularization and application. And the core challenge of high-dimensional constraint space optimization is solved. And glass research and development is pushed from trial and error experiments to calculation design.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of glass composition optimization, and particularly relates to a glass composition optimization method and system based on a neural network and a goat optimization algorithm. BACKGROUND

[0002] Glass is an important inorganic non-metallic material, widely used in the fields of building, automobile, electronics, optical devices, etc. The performance of glass (such as optical performance, mechanical performance, thermal stability, etc.) is mainly determined by its chemical composition. However, the relationship between glass composition and performance is highly complex and nonlinear, and traditionally relies on the "trial and error method" for composition optimization. This method needs to go through multiple iterations of "composition design-glass preparation-performance testing-composition optimization", which not only has a long experimental cycle (one melting-quenching process usually takes several hours or even several days), consumes a large amount of raw materials (a large amount of chemical raw materials are needed for each experiment), but also can only prepare a limited number of samples for each experiment, resulting in a very low efficiency and high cost in determining the optimal glass formula.

[0003] With the development of artificial intelligence technology, machine learning has shown great potential in the field of materials science. In recent years, researchers have used machine learning algorithms (such as neural networks, support vector machines, etc.) to establish prediction models from glass composition to performance. These models can quickly predict the performance of a given composition by learning from existing glass databases, thereby replacing some experiments in the virtual space.

[0004] However, current artificial intelligence technology research mainly focuses on "forward prediction", i.e. inputting the composition to predict the performance. In actual glass material design, the more critical requirement is "reverse design": according to specific performance requirements (such as high refractive index, low dispersion, specific thermal expansion coefficient, etc.), find a formula that meets the requirements in a huge composition space. Because the glass composition space is huge (usually containing multiple oxides, and each component ratio is continuously adjustable), and the performance constraints can be diverse (multi-objective, multi-constraint), traditional optimization methods (such as grid search, random search) are difficult to efficiently explore this high-dimensional space.

[0005] Therefore, how to use machine learning models to guide the reverse design of glass composition has become a difficult point in current research. Intelligent optimization algorithms (such as genetic algorithms, particle swarm optimization, ant colony algorithm, etc.) perform well in solving complex optimization problems, but their application in material design still faces challenges: On the one hand, optimization algorithms need to repeatedly call performance prediction models, and complex prediction models (such as deep neural networks) have high computational overhead; On the other hand, multi-objective optimization dilemma: glass composition design has strict constraint conditions, and the optimization algorithm needs to effectively handle these constraints. For example, when the transmittance is >92%, the thermal expansion coefficient is 8.5±0.2*10 -6 / K, the hardness is >6.5GPa, and other multi-constraint conditions are met, the success rate of the existing method is <35%.

[0006] In summary, the glass composition optimization system based on neural network and goat optimization algorithm is developed to overcome the low efficiency problem of the traditional trial and error method, solve the shortcomings of the existing machine learning method in reverse design, provide a new idea for the intelligent design of glass materials, and accelerate the research and development process of new glass materials. SUMMARY

[0007] The purpose of the present application is to provide a glass composition optimization method and system based on neural network and goat optimization algorithm, which solves the problem of the core challenge of high-dimensional constraint space optimization by the high-precision prediction ability of neural network and the global search and adaptive ability of goat optimization algorithm.

[0008] To solve the above technical problems, the present application is realized by the following technical scheme: As a first aspect provided by the present application, the present application is a glass composition optimization method based on neural network and goat optimization algorithm, comprising the following steps: Based on the glass composition data and the corresponding performance index data, training data set and validation data set are constructed; The neural network structure is constructed, and the activation function and the loss function are selected. The neural network is trained based on the training data set to obtain an initial neural network prediction model; The goat population parameters are initialized and substituted into the neural network prediction model. The performance index of the neural network prediction model is calculated based on the validation data set as the fitness value to construct the goat optimization algorithm framework; The goat optimization algorithm is used for iterative optimization, and the optimal parameters or glass composition combination obtained are taken as the optimization target to output the optimal neural network prediction model; The neural network prediction model is embedded into the goat optimization algorithm, and the glass composition is taken as the optimization variable. The performance index predicted by the neural network prediction model is taken as the fitness value to search and output the optimal glass composition combination.

[0009] Further, the glass composition data includes the content percentage of silicon oxide, sodium oxide, calcium oxide, aluminum oxide and magnesium oxide, and the performance index data includes one or more of refractive index, Abbe number, thermal expansion coefficient and chemical stability. The data is cleaned and normalized to construct the training data set and the validation data set.

[0010] Further, the neural network structure is a multi-layer perceptron (MLP), a convolutional neural network (CNN), or a recurrent neural network (RNN), the activation function is ReLU, Sigmoid, or Tanh, and the loss function is mean squared error (MSE) or cross-entropy loss.

[0011] Further, the method of initializing the goat herd parameters comprises: Initializing the goat herd: setting the size of the goat herd, the maximum number of iterations, and the dimension of the search space, and randomly initializing the positions of the goats as: Equation (1) where i = 1, 2,..., N, i is the number of individuals in the goat herd, and d is the dimension of the search space (i.e., the number of decision variables); Generation of the initial population: for each goat i (a total of N), generate a random position in each dimension j (a total of d dimensions): Equation (2) where UB is the upper bound vector, LB is the lower bound vector, j = 1...d, indicating that the search space has d dimensions, and rand(d) generates a d-dimensional random vector with values ranging between 0 and 1; The method of calculating the fitness value comprises: substituting the position of each goat (neural network hyperparameters or glass component combination) into the neural network model, and calculating the performance indicator of the neural network prediction model using the validation dataset as the fitness value.

[0012] Further, the goat herd parameters are set as: The size of the goat herd is set to 50-200, the maximum number of iterations is set to 200-1000, and the dimension of the search space is dynamically adjusted according to the optimization target.

[0013] Further, the method of iterative optimization using the goat optimization algorithm comprises: Updating the goat position: updating the position of each goat according to the foraging behavior, jumping mechanism, and group cooperation rules of the goat optimization algorithm; Boundary processing: if the goat position exceeds the boundary of the search space, it is pulled back to within the boundary; Recalculating the fitness: after each position update, the fitness value is recalculated; Iterative optimization: repeat the position update and fitness calculation process until the maximum number of iterations is reached or the fitness value meets the pre-set threshold, and obtain the optimal parameters or glass component combination.

[0014] Further, the method of updating the goat position comprises: Based on the foraging behavior, the goat moves to the area with abundant food, which simulates the exploration of the direction of high fitness solution, and the new position of each goat is updated as: Equation (3) Based on the jumping mechanism, the goat jumps a large range with a certain probability, escapes from the local optimum, and avoids falling into the local optimum. The goat position defined in the development stage is: Equation (4) The goat position defined in the development stage is updated as: Equation (5) Based on the poor performance of the parasitic solution and the generation of a new solution, if the fitness of the goat position is in the lowest 20% of the overall, the new position of each goat is updated as: Equation (6) wherein, is the position of the i-th goat at iteration t, a is an exploration coefficient, R is a random variable drawn from a Gaussian distribution N(0, 1), UB is an upper bound vector, LB is a lower bound vector, is the current optimal solution, J is a development coefficient; is the current optimal solution, is the development coefficient, and rand(d) generates a d-dimensional random vector with a value range of [0, 1].

[0015] Further, the stopping condition for iteration optimization using the goat optimization algorithm is: reaching the maximum number of iterations (Tmax); or the fitness improvement is lower than the predefined threshold : Equation (7) wherein, , are the positions of the goat at iteration t+1 and t, , are the fitnesses of , .

[0016] Further, it also includes a glass component correction method, and the correction method comprises: If the performance index predicted by the neural network prediction model is used as the fitness value, and the optimal glass component combination searched and output exceeds the actual feasible range (such as the oxide content being negative or exceeding 100%), then: truncation processing: forcibly set the glass component content exceeding the range to the boundary value; re-sampling: re-generate the glass component combination within a reasonable range.

[0017] As a second aspect provided by the present application, the present application is a glass composition optimization system based on a neural network and a goat optimization algorithm, which is used to implement the first glass composition optimization method.

[0018] The present application has the following beneficial effects: The present application significantly improves the efficiency and accuracy of glass composition design through the high-precision prediction capability of the neural network and the global search and adaptive capability of the goat optimization algorithm, provides an innovative solution for the research and development of high-performance glass materials, breaks through the fragmented predicament of glass composition "prediction-design", solves the core challenge of high-dimensional constraint space optimization, and pushes glass research and development from "trial and error experiment" to "computational design".

[0019] Of course, implementing any product of the present application does not necessarily need to achieve all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0021] Figure 1 Flow chart of the glass composition optimization method based on a neural network and a goat optimization algorithm of the present application. DETAILED DESCRIPTION

[0022] In the following description, specific details are set forth in order to provide a thorough understanding of embodiments of the application. However, persons of ordinary skill in the art will appreciate that embodiments of the application can be practiced without the specific details, other embodiments, and in other instances, well-known systems, structures, circuits, and methods have not been described in detail in order not to obscure the application.

[0023] It should be understood that when used in the specification and the appended claims, the term "comprises" indicates the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0024] It should also be understood that the term "and / or" used in the specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations thereof, and includes these combinations.

[0025] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0026] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] Example 1: like Figure 1 As shown, as the first embodiment provided by the present invention, the present invention is a glass composition optimization method based on neural networks and goat optimization algorithm, including the following steps: Based on glass composition data and corresponding performance index data, training datasets and validation datasets are constructed. Construct a neural network structure and select activation and loss functions. Train the neural network based on the training dataset to obtain an initial neural network prediction model. Initialize the goat herd parameters and substitute them into the neural network prediction model. Calculate the performance index of the neural network prediction model based on the validation dataset as the fitness value to construct the goat optimization algorithm framework. The optimal neural network prediction model is output by iteratively optimizing the goat optimization algorithm and taking the obtained optimal parameter or glass composition combination as the optimization objective. By embedding a neural network prediction model into a goat optimization algorithm, using glass composition as the optimization variable, and using the performance index predicted by the neural network prediction model as the fitness value, the optimal combination of glass composition is searched and output.

[0029] As an embodiment provided by the present application, preferably, the glass component data includes the content percentage of silicon oxide, sodium oxide, calcium oxide, aluminum oxide and magnesium oxide, the performance index data includes one or more of refractive index, Abbe number, thermal expansion coefficient and chemical stability, the data is cleaned and normalized, and a training data set and a verification data set are constructed.

[0030] As an embodiment provided by the present application, preferably, the neural network structure is a multi-layer perception (MLP), a convolutional neural network (CNN) or a recurrent neural network (RNN), the activation function is ReLU, Sigmoid or Tanh, and the loss function is mean square error (MSE) or cross-entropy loss.

[0031] As an embodiment provided by the present application, preferably, the goat herd parameter is set as: The goat herd size is set to 50-200, the maximum iteration number is set to 200-1000, and the search space dimension is dynamically adjusted according to the optimization target.

[0032] The performance comparison before and after the optimization of the goat optimization algorithm in Embodiment 1 is shown in Table 1. Table 1 R2 MSE MAE Before optimization 0.88 22.62 4.26 After optimization 0.95 3.47 1.79 Embodiment 2: As a second embodiment provided by the present application, the present application is a glass component optimization method based on neural network and goat optimization algorithm, including the following steps: Step (1): data collection and preprocessing: collect glass component data and its corresponding performance index data, clean and normalize the data, and construct a training data set and a verification data set; Step (2): constructing a neural network prediction model: designing a neural network structure, selecting an activation function and a loss function, training the neural network using the training data set in step (1), and obtaining an initial neural network prediction model; Step (3): constructing a goat optimization algorithm framework: Initializing the goat herd: setting the size of the goat herd, the maximum iteration number, the search space dimension (corresponding to the neural network hyperparameters or glass component variables), and randomly initializing the position of the goat as: Formula (1) Wherein, i=1,2,…,N, i is the number of individuals in the goat herd, d is the search space dimension (i.e. the number of decision variables); Generating the initial population, for each goat i (a total of N), generating a random position in each dimension j (a total of d dimensions): Formula (2) Wherein, UB is the upper boundary vector, LB is the lower boundary vector, j = 1...d, represents that the search space is d-dimensional, and rand(d) generates a d-dimensional random vector with a value range of [0, 1]; The method for calculating the fitness value comprises the following steps: the position (neural network hyperparameter or glass component combination) of each goat is substituted into the neural network model, and the performance index of the neural network prediction model is calculated by using the verification data set as the fitness value.

[0033] Step (4): optimizing the neural network hyperparameter or glass component by using the goat optimization algorithm: The method for iterative optimization by using the goat optimization algorithm comprises the following steps: Updating the position of the goat: the position of each goat is updated according to the foraging behavior, jumping mechanism and group cooperation rule of the goat optimization algorithm; Boundary processing: if the position of the goat exceeds the boundary of the search space, it is pulled back to the boundary; Recalculating the fitness value: the fitness value is recalculated after each position update; Iterative optimization: the position updating and fitness value calculating process is repeated until the maximum iteration number is reached or the fitness value meets the preset threshold, and the optimal parameter or glass component combination is obtained.

[0034] Step (5): collaborative optimization process: Phase one: taking the neural network hyperparameter as the optimization target, the optimal neural network prediction model is obtained by using the goat optimization algorithm; Phase two: embedding the optimized neural network prediction model into the goat optimization algorithm, taking the glass component as the optimization variable, and searching for the optimal glass component combination by using the neural network prediction performance index as the fitness value.

[0035] Step (6) outputting the optimal result: outputting the optimal neural network prediction model parameter in phase one or the optimal glass component combination in phase two.

[0036] Embodiment 3: As a third embodiment provided by the present application, the present application is a glass component optimization method based on a neural network and a goat optimization algorithm, based on embodiment 1 or 2, the method for updating the position of the goat comprises the following steps: Based on the foraging behavior, the goat moves to the food-rich area, which is simulated to explore in the direction of the high fitness solution, in this stage, each goat is inspired by the foraging behavior, and moves randomly in different directions to explore the search space, and the new position of each goat is updated as: Formula (3) Based on the jumping mechanism, the goat jumps a large range with a certain probability, escapes from the local optimum, avoids falling into the local optimum, and gradually moves to the best solution so far, simulating the trend of the goat moving to the best grazing area; the goat position defined in the development stage is: Equation (4) The sudden jump is used to escape from the local optimum, and the goat position defined in the development stage is updated based on the unique movement ability of the goat in rugged terrain: Equation (5) In order to maintain the quality of the scheme, the goat avoids the area infected by parasites, eliminates the poor solution based on the performance of the parasites, and generates a new solution, if the fitness of the is in the lowest 20% of the overall, the new position of each goat is updated as: Equation (6); Wherein, is the position of the i-th goat at iteration t, alpha is the exploration coefficient, R is a random variable drawn from a Gaussian distribution N(0,1), UB is the upper boundary vector, LB is the lower boundary vector, is the current optimal solution, J is the development coefficient; is the current optimal solution, is the development coefficient, rand(d) generates a d-dimensional random vector with a value range of [0,1]; the diversity and robustness are enhanced, and stagnation in a poor local optimum state is prevented.

[0037] Embodiment 4: As the fourth embodiment provided by the application, the application is a glass composition optimization method based on neural network and goat optimization algorithm, based on embodiment 1 or 2 or 3, the stop condition of iterative optimization using the goat optimization algorithm is: The maximum iteration number (Tmax) is reached; Or the fitness improvement is lower than the predefined threshold : Equation (7) Wherein, , are the positions of the goat at iteration t+1 and t respectively, , are the fitnesses of , respectively.

[0038] Embodiment 5: As the fifth embodiment provided by the present application, the present application is a glass component optimization method based on neural network and goat optimization algorithm, based on embodiment 1 or 2 or 3 or 4, further comprising a glass component correction method, the correction method comprising: If the performance index predicted by the neural network prediction model is used as the fitness value, the optimal glass component combination searched and output exceeds the actual feasible range (such as the oxide content being negative or exceeding 100%), then: Truncation processing: the glass component content exceeding the range is forced to be set as the boundary value; Resampling: regenerate the glass component combination within a reasonable range.

[0039] Embodiment 6: As the sixth embodiment provided by the present application, the present application is a glass component optimization method based on neural network and goat optimization algorithm, based on embodiment 1 or 2 or 3 or 4 or 5, when the glass component is optimized based on neural network and goat optimization algorithm, the collaborative optimization engine supports the following two modes: Staged optimization mode: first optimize the neural network hyperparameters, then optimize the glass component; Joint optimization mode: optimize the neural network parameters and the glass component at the same time, and improve the performance of both by alternating iteration.

[0040] Embodiment 7: As the seventh embodiment provided by the present application, the present application is a glass component optimization system based on neural network and goat optimization algorithm, which is used to realize the glass component optimization method described in embodiments 1-6.

[0041] Embodiment 8: The present application is a glass component optimization method based on neural network and goat optimization algorithm, comprising the following steps: Step (1): data collection and preprocessing: collect the elastic modulus data of different component oxide glasses, construct a glass elastic modulus database, specifically collect the elastic modulus data of 1000 groups of SiO2-Na2O-CaO-Al2O3 glasses with different contents, construct a glass elastic modulus database, and clean and normalize the data to construct a training data set and a validation data set; Step (2): construct a neural network prediction model: design a neural network structure, select an activation function and a loss function, train the neural network using the training data set in step (1), and obtain an initial neural network prediction model; Step (3) construct a goat optimization algorithm framework: Initialize the goat herd: set the size of the goat herd to 100, the maximum number of iterations to 1000, the search space dimension (neural network hyperparameters), and randomly initialize the position of the goat; (1) where i = 1, 2, …, N, d is the number of decision variables, and the formula for generating the initial population is: (2) where UB is the upper bound, LB is the lower bound, and rand(d) generates a d-dimensional random vector with values ranging between [0, 1].

[0042] Calculate the fitness: substitute the position of each goat (neural network hyperparameters) into the neural network prediction model, and calculate the model performance indicators using the validation dataset as the fitness value; Step (4): optimize the neural network hyperparameters using the goat optimization algorithm: Update the goat position: update the position of each goat according to the foraging behavior, jumping mechanism and group cooperation rules of the goat optimization algorithm; Boundary processing: if the goat position exceeds the search space boundary, pull it back to the boundary; Recalculate the fitness: after each position update, recalculate the fitness value; Iterative optimization: repeat the position update and fitness calculation process until the maximum iteration number is reached or the fitness value meets the preset threshold, and obtain the optimal parameters or component combination; Step (5) collaborative optimization process: Phase one: optimize the neural network hyperparameters to obtain the optimal neural network prediction model using the goat optimization algorithm; Phase two: embed the optimized neural network prediction model into the goat optimization algorithm, take the glass composition as the optimization variable, and search for the optimal glass composition combination by using the neural network prediction performance indicators as the fitness value; Step (6) output the optimal result: output the optimal neural network prediction model parameters in phase one or the optimal glass composition combination in phase two.

[0043] Example 9: The present application is a glass composition optimization method based on neural network and goat optimization algorithm, which includes the following steps: Step (1): data collection and preprocessing: collect the elastic modulus data of different component oxide glasses, construct a glass elastic modulus database, specifically collect the transmittance data of 2000 groups of SiO2-Na2O-CaO-Al2O3-MgO glasses with different contents, construct a glass elastic modulus database, and perform cleaning and normalization processing on the data to construct a training dataset and a validation dataset; Step (2): construct a neural network prediction model: design the neural network structure, select the activation function and loss function, train the neural network using the training dataset in step (1), and obtain the initial neural network prediction model; Step (3): Building the framework of the goat optimization algorithm: Initialize the goat herd: Set the size of the goat herd to 150, the maximum number of iterations to 2000, the search space dimension (neural network hyperparameters), and randomly initialize the positions of the goats. (1) where i = 1, 2, …, N, d is the number of decision variables (dimensions), and the generation formula for the initial population is: (2) where UB is the upper bound, LB is the lower bound, and rand(d) generates a d-dimensional random vector with values ranging between 0 and 1.

[0044] Calculate the fitness: Substitute the position of each goat (neural network hyperparameters) into the neural network prediction model, and use the validation dataset to calculate the model performance indicators as the fitness value. Step (4): Optimize the neural network hyperparameters using the goat optimization algorithm: Update the goat position: Update the position of each goat according to the foraging behavior, jumping mechanism, and group cooperation rules of the goat optimization algorithm. Boundary processing: If the goat position exceeds the search space boundary, pull it back to within the boundary. Recalculate the fitness: Recalculate the fitness value after each position update. Iterative optimization: Repeat the position update and fitness calculation process until the maximum number of iterations is reached or the fitness value meets the pre-set threshold, obtaining the optimal parameters or component combination. Step (5): Collaborative optimization process: Phase one: Optimize the neural network hyperparameters using the goat optimization algorithm to obtain the optimal neural network prediction model. Phase two: Embed the optimized neural network prediction model into the goat optimization algorithm, use the glass composition as the optimization variable, and use the neural network prediction performance indicators as the fitness value to search for the optimal glass composition combination. Step (6): Output the optimal result: Output the optimal neural network prediction model parameters in phase one or the optimal glass composition combination in phase two.

[0045] Comparative Example 1: The same 1000 sets of different content SiO2-Na2O-CaO-Al2O3 glass elastic modulus data as in Example 8 were used to build a neural network prediction model. However, in Comparative Example 1, the neural network hyperparameters were empirically set, the glass composition optimization used grid search and did not use the goat optimization algorithm for optimization, and the other steps were the same as in Example 8.

[0046] Table 2: Optimization performance of goat optimization algorithm and neural network prediction model corresponding to examples 8, 9, and comparative example 1 R2 MSE MAE Comparative Example 1 0.89 21.32 3.05 Example 8 0.95 3.67 1.89 Example 9 0.96 4.38 1.95 wherein R 2 is the coefficient of determination used to measure the proportion of variation in the observed data explained by the model; MSE is the mean squared error, which is the average of the squared differences between predicted and actual values; MAE is the mean absolute error, which is the average of the absolute values of the differences between predicted and actual values.

[0047] From Table 2, we can see that the traditional modeling method (comparative example 1) that deviates from intelligent optimization algorithm has fundamental limitations: The hyperparameters depend on empirical setting, and the model accuracy is limited; Grid search dimension disaster, high-dimensional component space exploration failure; Single-objective optimization, unable to meet the industrial multi-index demand; And the present application based on neural network and goat optimization algorithm collaborative optimization, improves the model precision, R 2 improves 7.9%, realizes efficient component reverse design, and provides an engineeringable intelligent solution for glass new material research and development.

[0048] The glass component optimization method and system based on neural network and goat optimization algorithm significantly improves the efficiency and accuracy of glass component design through the high-precision prediction ability of neural network and the global search and adaptive ability of goat optimization algorithm, providing an innovative solution for the research and development of high-performance glass materials, breaking through the fragmented dilemma of glass component "prediction-design", and solving the core challenge of high-dimensional constraint space optimization. Glass research and development is pushed from "trial and error experiment" to "computational design".

[0049] In the description of the present specification, the description referring to the terms "one embodiment", "example", "specific example" and the like means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0050] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments described. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to best explain the principles of the application and its practical application to thereby enable others skilled in the art to best utilize the application and get the best results from the application. The application is only limited by the claims and their full scope and equivalents.

Claims

1. A glass composition optimization method based on neural network and goat optimization algorithm, characterized in that, The method comprises the following steps: constructing a neural network structure and selecting an activation function and a loss function, training the neural network based on a training data set, and obtaining an initial neural network prediction model; initializing goat herd parameters and substituting them into the neural network prediction model, calculating the performance indicators of the neural network prediction model based on a verification data set as fitness values to construct a goat optimization algorithm framework; iterative optimization using the goat optimization algorithm, and taking the optimal parameters or glass component combination as the optimization target to output an optimal neural network prediction model; embedding the neural network prediction model in the goat optimization algorithm, taking glass components as optimization variables, and searching for and outputting an optimal glass component combination by predicting the performance indicators of the neural network prediction model as fitness values.

2. The glass composition optimization method based on neural network and goat optimization algorithm according to claim 1, characterized in that, It also includes: constructing a training data set and a verification data set based on glass component data and corresponding performance indicator data; The glass component data includes the content percentage of silicon oxide, sodium oxide, calcium oxide, aluminum oxide and magnesium oxide, and the performance indicator data includes one or more of refractive index, Abbe number, thermal expansion coefficient and chemical stability. 3.The glass composition optimization method based on neural network and goat optimization algorithm according to claim 1, wherein, The neural network structure is a multilayer perceptron, a convolutional neural network or a recurrent neural network, the activation function is ReLU, Sigmoid or Tanh, and the loss function is mean square error or cross-entropy loss. 4.The glass composition optimization method based on neural network and goat optimization algorithm according to claim 1, wherein, The method for initializing the goat herd parameters comprises: initializing the goat herd: setting the size of the goat herd, the maximum number of iterations, and the dimension of the search space, and randomly initializing the position of the goat; initial population generation: for each goat, a random position is generated in each dimension.

5. The glass composition optimization method based on neural network and goat optimization algorithm according to claim 4, characterized in that, The initialized goat herd parameters are: The size of the goat herd is set to 50-200, the maximum number of iterations is set to 200-1000, and the dimension of the search space is dynamically adjusted according to the optimization target. 6.The glass composition optimization method based on neural network and goat optimization algorithm according to claim 1, wherein, The method for iterative optimization using the goat optimization algorithm comprises: updating the position of the goat: updating the position of each goat according to the foraging behavior, jumping mechanism and group cooperation rules of the goat optimization algorithm; boundary processing: if the position of the goat exceeds the boundary of the search space, it is pulled back to the boundary; recalculating the fitness: after each position update, the fitness value is recalculated; iterative optimization: repeat the position update and fitness calculation process until the maximum number of iterations is reached or the fitness value meets the pre-set threshold, and obtain the optimal parameters or glass component combination.

7. The glass composition optimization method based on neural network and goat optimization algorithm according to claim 6, characterized in that, The method for updating the position of the goat comprises: based on the foraging behavior: the goat moves to the area rich in food, simulating the exploration of the direction of the high fitness solution, and updating the position of each goat; based on the jumping mechanism: the goat jumps in a large range with a certain probability, and uses the jump to escape from the local optimum, and updates the position of the goat defined in the development stage; based on the parasite: eliminate the poor solution and generate a new solution, if the fitness of the position of the goat is in the lowest 20% of the overall, update the new position of each goat. 8.The glass composition optimization method based on neural network and goat optimization algorithm according to claim 1, wherein, The stopping condition for iterative optimization using the goat optimization algorithm is: reach the maximum number of iterations; or the fitness improvement is lower than the pre-defined threshold. 9.The glass composition optimization method based on neural network and goat optimization algorithm according to claim 1, wherein, The glass component correction method further comprises: if the optimal glass component combination searched and output by the neural network prediction model as fitness values exceeds the actual feasible range, then: Truncation: Force out-of-range glass composition content to the boundary value; Resampling: Regenerate the glass composition combination within a reasonable range.

10. A glass composition optimization system based on neural network and goat optimization algorithm, characterized in that, The glass composition optimization system is used to implement the glass composition optimization method of any one of claims 1-9.

Images