Composite membrane performance parameter optimization method based on neural network

Abstract

The invention provides a composite membrane performance parameter optimization method based on a neural network, which relates to the technical field of composite membranes, and is used for dynamically regulating and controlling the optimization priority of targets by setting an optimization target set; a two-stage optimization structure is constructed, a first layer adopts neural network input layer additional weight coding, and a second layer automatically adjusts a hidden layer activation function based on target sensitivity to adapt to different optimization requirements; a dynamic error compensation mechanism is adopted, conjoint analysis is carried out on target errors, a neural network loss function is adjusted, and the errors are controlled in different optimization modes; through continuous two rounds of optimization iteration, the target weight is adjusted, and the weight code is corrected according to the error change trend, so that the adaptive switching of the optimization target is realized, and the comprehensive performance of the composite membrane is improved. According to the method, higher optimization precision and adaptability can be achieved in the aspect of performance parameter optimization of the composite membrane, and the comprehensive performance of the composite membrane is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of composite membranes, and more specifically, to a composite membrane performance parameter optimization method based on neural network. Background Art

[0002] With the development of advanced manufacturing technology, the preparation process of composite membranes has been continuously improved, so that the material performance has been improved in terms of transmittance, mechanical strength, durability, etc. However, since composite membranes usually involve the coordinated optimization of multiple physical performance parameters, how to strike a balance between different performance indicators has become an important research direction for the current optimization design of composite membranes. Traditional optimization methods mainly rely on empirical parameter adjustment or optimization models based on experimental data, which have many shortcomings. On the one hand, most optimization methods fail to fully consider the complex coupling relationship between different performance parameters, and local optimal problems are prone to occur during the optimization process. For example, while optimizing transmittance, mechanical strength often decreases, and the existing methods lack a mechanism for dynamically adjusting the optimization weights, which limits the optimization effect. On the other hand, the adaptive ability of existing optimization methods is weak, and most of them use fixed parameters or a single optimization strategy, which is difficult to adjust in real time according to different application requirements. In addition, although the traditional neural network optimization method can improve the performance of composite membranes to a certain extent, due to the fixed key parameters such as activation function and loss function, it is difficult to adjust the optimization target adaptively with the change of error, and the optimization efficiency is low. Therefore, there is an urgent need for an optimization method that can comprehensively consider multi-objective optimization requirements and has dynamic regulation capabilities to improve the comprehensive performance of composite membranes. Summary of the invention

[0003] In order to solve the above technical problems, the present invention is proposed. The present invention provides a composite membrane performance parameter optimization method based on neural network.

[0004] According to one aspect of the present invention, a method for optimizing performance parameters of a composite membrane based on a neural network is provided, which includes: setting an optimization target set for dynamically adjusting the optimization priority of the target; constructing a two-level optimization structure, in which the first layer uses a neural network input layer with additional weight coding, and the second layer automatically adjusts the hidden layer activation function based on the target sensitivity to adapt to different optimization needs; using a dynamic error compensation mechanism to jointly analyze the target errors and adjust the neural network loss function to control the error under different optimization modes; adjusting the target weights through two consecutive rounds of optimization iterations, and correcting the weight coding according to the error change trend to achieve adaptive switching of optimization targets and improve the comprehensive performance of the composite membrane.

[0005] Furthermore, the objectives include light transmittance and mechanical strength.

[0006] Furthermore, the optimization target set is ,in, is the transmittance target, For mechanical strength purposes.

[0007] Furthermore, the optimization priorities of the dynamic control objectives include: Setting optimization priority parameters , control the weights of transmittance and mechanical strength in the optimization process; perform adaptive target weight adjustment based on experimental data; and identify the target weight configuration used in the current optimization process by optimizing the target encoding.

[0008] Furthermore, the two-level optimization structure includes: first-layer optimization: by adding optimization target weight encoding to the neural network input layer, the network can understand the current optimization priority and distinguish different optimization requirements; second-layer optimization: using a target sensitivity mechanism, according to the degree of influence of the optimization target on the result, the hidden layer activation function is adaptively adjusted, so that the network can adapt to different optimization requirements.

[0009] Furthermore, the second-layer optimization includes: using the optimization target set G to derive the composite membrane structure parameters to express the degree of influence of changes in different composite membrane structure parameters on the target optimization, that is, the target sensitivity; establishing a target sensitivity matrix according to the target sensitivity; and adjusting the activation function of the hidden layer according to the calculation results of the target sensitivity matrix.

[0010] Furthermore, the dynamic error compensation mechanism is used to jointly analyze the target errors, including: calculating the transmittance error and the mechanical strength error respectively; ensuring the correctness of the optimization direction according to the error ratio of the transmittance error and the mechanical strength error; and calculating the dynamic error compensation factor to dynamically adjust the optimization weights of the transmittance error and the mechanical strength error.

[0011] Furthermore, the transmittance error is measured by the difference between the predicted value of the neural network and the actual value measured experimentally; the mechanical strength error is uniformly calculated by comparing the predicted value with the actual value.

[0012] Further, the adjustment of the neural network loss function to control the error in different optimization modes includes: the optimization modes include a transmittance priority mode, a mechanical strength priority mode, and a balanced optimization mode; in different optimization modes, the loss function is dynamically adjusted, including: in the transmittance priority mode, the loss function increases the weight of the transmittance error; in the mechanical strength priority mode, the loss function increases the weight of the mechanical strength error; In the balanced optimization mode, the loss function balances the transmittance error and the mechanical strength error so that the optimization weights of the two remain stable.

[0013] Furthermore, the two consecutive rounds of optimization iterations include: in the first round of optimization, reading the optimization target weight code, calculating the error compensation factor β, and dynamically adjusting the optimization target weight based on the error change trend to ensure the balance of the optimization strategy; in the second round of optimization, using the optimization target weight adjusted in the first round for training, and continuously dynamically adjusting the error compensation factor β to optimize the error balance, and calculating the error convergence. If the error change slows down, the weight is further adjusted, otherwise the optimization convergence is judged to ensure that the final optimization direction is correct.

[0014] Compared with the prior art, the method and method for optimizing composite membrane performance parameters based on neural network provided by the present invention, through the first layer, by adding optimization target weight coding to the neural network input layer, enables the network to identify different optimization requirements and adjust the optimization strategy accordingly; the second layer adaptively adjusts the hidden layer activation function based on the target sensitivity to improve the adaptability of the neural network to different optimization targets. In addition, the introduction of the dynamic error compensation mechanism ensures that the transmittance and mechanical strength errors are balanced during the optimization process to avoid excessive bias of a certain optimization target, resulting in a decrease in overall performance. Two consecutive rounds of optimization iterations further improve the adaptability of the optimization strategy, and adjust the target weight coding through the error change trend, making the optimization process more intelligent and avoiding falling into a local optimal solution. The present invention can achieve higher optimization accuracy and adaptability in the optimization of composite membrane performance parameters, and improve the comprehensive performance of the composite membrane. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 The present invention is a flowchart of a composite membrane performance parameter optimization method based on a neural network according to an embodiment of the present invention.

[0016] Figure 2 It is a schematic diagram of the optimization priority of the dynamic control target in the neural network-based composite membrane performance parameter optimization method according to an embodiment of the present invention.

[0017] Figure 3 Schematic diagram of a dual-stage optimization structure in a neural network-based composite membrane performance parameter optimization method according to an embodiment of the present invention.

[0018] Figure 4 Schematic diagram of two rounds of optimization iterations in a neural network-based composite membrane performance parameter optimization method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] Below, the exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention, and it should be understood that the present invention is not limited to the exemplary embodiments described here.

[0020] As mentioned in the background technology above, since composite membranes usually involve the coordinated optimization of multiple physical performance parameters, how to strike a balance between different performance indicators has become an important research direction for the current optimization design of composite membranes. Traditional optimization methods mainly rely on empirical parameter adjustment or optimization models based on experimental data, such as response surface analysis (RSM), genetic algorithm (GA) and finite element analysis (FEA). These methods can improve certain properties of composite membranes to a certain extent, but due to the lack of precise adjustment of multi-objective trade-offs during the optimization process, while improving a certain performance, other key performances may be sacrificed, making it difficult to meet the needs of high-precision and intelligent optimization.

[0021] Existing composite membrane optimization technologies have many shortcomings. On the one hand, most optimization methods fail to fully consider the complex coupling relationship between different performance parameters, and local optimal problems are prone to occur during the optimization process.

[0022] Therefore, a neural network-based optimization scheme for composite membrane performance parameters is needed.

[0023] Figure 1 FIG. 1 is a flow chart of a method for optimizing composite membrane performance parameters based on a neural network according to an embodiment of the present invention. Figure 1 As shown, the composite membrane performance parameter optimization method based on neural network includes the following steps: S1: Set a set of optimization targets to dynamically control the optimization priority of the targets; S2: Construct a two-level optimization structure, the first layer uses the neural network input layer to add weight coding, and the second layer automatically adjusts the hidden layer activation function based on the target sensitivity to adapt to different optimization needs; S3: Adopt a dynamic error compensation mechanism to jointly analyze the target errors and adjust the neural network loss function to control the error under different optimization modes; S4: Through two consecutive rounds of optimization iterations, adjust the target weights, and correct the weight coding according to the error change trend to achieve adaptive switching of optimization targets and improve the comprehensive performance of the composite membrane.

[0024] In the production and application of composite films, light transmittance and mechanical strength are the two most important performance indicators, and there is usually a conflict between the two. However, these two goals have a certain conflict. For example, improving light transmittance often means reducing the thickness of the film or changing the material ratio, which may lead to a decrease in mechanical strength; conversely, improving mechanical strength may affect the light transmittance of the material. Therefore, it is difficult for traditional single-objective optimization methods to find the optimal balance between the two goals, which easily leads to optimization biased towards a certain characteristic and failure to take into account the overall performance. Therefore, how to find a suitable optimization balance between the two is the core problem that the present invention needs to solve.

[0025] Traditional optimization methods often adopt a single-objective optimization strategy, that is, optimizing transmittance or mechanical strength separately, and cannot flexibly adjust the target priority under different demand scenarios. In addition, the traditional optimization method lacks a data-driven intelligent control mechanism, which makes the optimization direction easily deviate from the actual application requirements. This step S1 sets an optimization target set and introduces a mechanism for dynamically controlling the optimization priority, so that the optimization process can adapt to different application requirements and provide reasonable target settings for subsequent optimization strategies. In an embodiment of the present invention, the step S1 specifically includes: In order to construct a reasonable set of optimization targets, it is first necessary to collect the performance data of the composite film, including light transmittance and mechanical strength. The specific method of data collection is as follows: Using a UV-visible spectrophotometer, the composite film sample was fixed on an optical test stand, and the sample was illuminated by a light source. The transmittance was calculated by detecting the intensity of the transmitted light, and the transmittance at different wavelengths was recorded, and the average transmittance was determined. Using an electronic universal testing machine, the composite film was subjected to a tensile test in accordance with the ASTM D882 standard, the maximum tensile strength (MPa) was recorded, the tensile strength of multiple samples was measured, and the average value was determined. The 3σ principle was used to eliminate abnormal experimental data to avoid measurement errors affecting the optimization process, and the transmittance and mechanical strength were normalized to ensure that the data was within the range of 0-1, making the neural network training more stable.

[0026] Set optimization goal set ,in, It is the light transmittance target, which is used to measure the light transmittance of the composite film. Its unit is percentage (%). The higher the value, the better the light transmittance performance. It is a mechanical strength target, used to measure the tensile and tear resistance of the composite film. Its unit is MPa (megapascal). The higher the value, the better the mechanical strength.

[0027] Optional, optimization target set The setting mode makes this method suitable for many different application scenarios, such as: Display screen film: Prioritize light transmittance ; Protective film material: Prioritize mechanical strength ; Comprehensive application of membrane materials: balance and .

[0028] Furthermore, after setting the optimization target set, such as Figure 2 As shown, it is necessary to define the optimization priority , which enables the optimization process to flexibly adjust between different objectives: Optimizing priority parameters , which is used to control the weights of transmittance and mechanical strength during the optimization process and meets the following conditions: , only optimize the transmittance , ignoring the mechanical strength ; 0, optimize only mechanical strength , ignoring the transmittance ; , transmittance and mechanical strength are equally optimized, that is, the balance mode.

[0029] The better one is to use the adaptive target weight adjustment method based on experimental data to calculate Set the average transmittance and average mechanical strength in the experimental data, and calculate the transmittance and mechanical strength of the current film material. The calculation formula is: in, and are the transmittance and mechanical strength of the current film material, and are the average transmittance and average mechanical strength in the experimental data, respectively.

[0030] If the transmittance of the current film material deviates greatly from the historical average, the transmittance will be given priority in the optimization process, that is, Bias 1: If the mechanical strength of the current membrane material deviates greatly from the historical mean, the mechanical strength will be enhanced first during the optimization process, that is, biased towards 0; if the two deviations are close, then Close to 0.5, achieving balanced optimization.

[0031] In order to enable the neural network to process the optimization target set G, the present invention introduces the optimization target encoding, which is used to identify the target weight configuration used in the current optimization process: 01 means >0.5, that is, the transmittance is optimized first; 10 means <0.5, that is, mechanical strength is optimized first; 11 means = 0.5, that is, the two are balanced and optimized. This encoding method can be used in the input layer of the neural network to make the regulation of the optimization strategy more intelligent and structured.

[0032] It should be noted that there is a contradiction between the transmittance and mechanical strength targets. This step dynamically adjusts the optimization priority to make the optimization direction more in line with actual needs and adjusts the weights adaptively. , reducing unnecessary iterations during training and improving the convergence speed of the optimization algorithm. Optimization target encoding enables the neural network to adapt to different types of film materials and improves the scope of application of the optimization method. Compared with linear weight allocation, which requires manual weight setting and lacks adaptability, the multi-objective optimization method of the present invention adopts a data-driven dynamic control mechanism, which can automatically adjust the optimization target priority, taking into account both transmittance and mechanical strength, and improving optimization efficiency, and is irreplaceable.

[0033] In the embodiment of the present invention, the step S2 needs to be based on the optimization target setting of S1, and a two-level optimization structure adapted to different optimization requirements is constructed, so that the neural network can more accurately optimize the performance parameters of the composite membrane, such as Figure 3 As shown in the figure, it specifically includes the first layer optimization (input layer optimization): by adding optimization target weight encoding to the input layer of the neural network, the network can understand the current optimization priority and distinguish different optimization requirements; the second layer optimization (hidden layer optimization): using the target sensitivity mechanism, according to the degree of influence of the optimization target on the result, the hidden layer activation function is adaptively adjusted, so that the network can adapt to different optimization requirements and improve the optimization accuracy and stability.

[0034] Specifically, the first layer optimization includes: in order to enable the neural network to identify the current optimization priority (transmittance priority, mechanical strength priority, or a balance between the two), it is necessary to add a set of optimization target weight codes to the input layer. The optimization target weight codes come from the optimization target priority set in step S1. The input data of the neural network includes the following two parts: Structural parameters of the composite membrane: including membrane thickness, porosity, material ratio and other key parameters affecting the performance of the composite membrane and optimization target weight coding: optimization priority parameters from S1 , and add the input layer in an encoded way. The input data format is as follows: ,in, represents the structural parameters of the composite membrane, represents the total number of structural parameters of the composite film, that is, the dimension of the input feature. , represents the weight encoding of the optimization target, which is calculated by the contribution of the optimization target itself. It can be seen that the influence of the optimization target is no longer discrete, but continuous, which allows the neural network to adjust the optimization direction more smoothly and improve the optimization accuracy.

[0035] During the optimization process, the normalized values ​​of transmittance and mechanical strength were calculated to ensure that their dimensions were consistent so that they could be used in the calculation. When it has a reasonable numerical range, then calculate the optimized target weight encoding ,in, The value range of is [0,1], indicating the current optimization focus: , which means that transmittance optimization is prioritized (i.e. Take the larger value); when This means that mechanical strength optimization takes precedence (i.e. Take the smaller value); when , it indicates that both are optimized in a balanced manner.

[0036] It should be noted that by introducing With this continuous variable, the neural network can adjust the optimization strategy more smoothly instead of relying on discrete binary coding, thereby improving the optimization accuracy. This weight encoding method can be directly input into the neural network without the need for additional logical judgment, making the optimization process more efficient.

[0037] Specifically, the second layer optimization includes: the hidden layer of the neural network is responsible for learning the complex features in the data and adjusting the optimization direction. In this step, the target sensitivity analysis mechanism is introduced to adaptively adjust the activation function of the hidden layer according to the influence of transmittance and mechanical strength on the optimization results to improve the optimization accuracy and stability.

[0038] The target sensitivity is used to measure the influence of the optimization target on the optimization result. The optimization target set G can be used to derive the composite membrane structure parameters to indicate the influence of the change of the composite membrane structure parameters on the target optimization.

[0039] Furthermore, in order to facilitate the control of the optimization process, a target sensitivity matrix is ​​constructed : in, Indicates the parameter for transmittance sensitivity; Indicative parameters for mechanical strength According to the target sensitivity matrix The calculation results of the hidden layer are used to adjust the activation function so that the neural network can better adapt to the current optimization needs. The specific strategy is as follows: If the transmittance Larger (i.e., transmittance has a greater impact on the optimization results), the ReLU activation function is used to improve the stability of transmittance optimization; if the mechanical strength If the sensitivity of the two is similar, the Sigmoid activation function is used to make the optimization process more balanced. This mechanism enables the neural network to automatically adapt to different optimization objectives and improve the reliability of the optimization results.

[0040] It should be noted that the present invention enables the neural network to adjust the optimization strategy according to different optimization requirements by adding optimization target coding to the input layer; adaptively adjusts the hidden layer activation function to make the optimization process more stable and reduce the computational cost; and through target sensitivity analysis, ensures the applicability of the optimization results under different composite membrane structures.

[0041] In the embodiment of the present invention, although a two-level optimization structure is constructed in the above process, in which the first layer (input layer) is used as the neural network input through additional weight encoding, and the second layer (hidden layer) uses target sensitivity analysis to adaptively adjust the activation function to adapt to different optimization needs, however, the neural network is usually affected by the following problems during the optimization process: for example, the competitive relationship between the optimization targets: there is a mutual constraint relationship between transmittance and mechanical strength, and simply optimizing one indicator may lead to the deterioration of another indicator; due to factors such as data noise, measurement error, parameter uncertainty, etc., the prediction results of the neural network may gradually deviate from the optimal solution, resulting in difficulty in model convergence or optimization failure; and under different optimization modes (transmittance priority, strength priority, balanced optimization), the weight of the impact of the error on the final optimization result is different, and the use of a fixed loss function cannot take into account all situations.

[0042] In traditional neural network optimization methods, the loss function is often fixed, such as mean square error (MSE) or cross entropy. However, in multi-objective optimization tasks, a single loss function is difficult to effectively handle the balance problem between objectives. For example, using only MSE may result in a good optimization effect of transmittance, but too much loss of mechanical strength, and using only Huber loss may improve robustness, but it is impossible to adjust the error weight according to the optimization mode. Therefore, the present invention introduces a dynamic error compensation mechanism, and the step S3 specifically includes: During the training process, the transmittance error and mechanical strength error are calculated separately to determine the focus of the current optimization. The transmittance error can be measured by the difference between the predicted value of the neural network and the actual value measured experimentally. The size of the error determines the accuracy of the transmittance optimization. The mechanical strength error can also be calculated by comparing the predicted value with the actual value. The size of the error reflects the reliability of the optimization algorithm in predicting the mechanical strength.

[0043] To ensure that the optimization direction is correct, it is necessary to determine the error ratio of the two, that is: if the transmittance error is much larger than the mechanical strength error, the optimization weight of the transmittance needs to be increased to reduce the transmittance deviation; if the mechanical strength error is much larger than the transmittance error, it is necessary to adjust the optimization strategy to focus on mechanical strength; if the errors of the two are close, a balanced optimization strategy should be adopted to reduce the errors of both at the same time.

[0044] In addition, the relative importance of errors will vary in different optimization modes: in transmittance priority mode, the mechanical strength error is allowed to increase appropriately, but the transmittance error must be strictly controlled; in mechanical strength priority mode, the transmittance error can increase appropriately, but the mechanical strength error must be reduced to the lowest level; in balanced optimization mode, both errors need to be strictly controlled to ensure the stability of the overall performance.

[0045] Furthermore, in order to further optimize the error compensation, the dynamic error compensation factor is calculated To dynamically adjust the optimization weights of transmittance error and mechanical strength error: in, is a normalization factor used to avoid the denominator being zero and to enhance numerical stability. Encode the optimization target weight for transmittance, Encode the optimization objective weights for mechanical strength; It can be seen that The calculation is the relative difference between the weight encodings of the two optimization objectives (transmittance and mechanical strength). The current weighting method usually uses fixed weights, and introduces dynamic error compensation factors. Afterwards, the present invention hopes to enable the weight factor to be dynamically adjusted to meet the needs of different optimization objectives. Therefore, the weight factor is expressed as: when →0, that is, the optimization weights of the two are close, and the weight distribution keeps the original weight unchanged; when →1, the weight distribution will tilt towards the direction with smaller error to prevent a single error from being too large and affecting the overall optimization. This method can ensure the dynamic balance between optimization objectives and make the error distribution more reasonable.

[0046] It can be seen that when the transmittance error is large, the error compensation factor will automatically increase the optimization weight of the transmittance, so that the neural network pays more attention to the transmittance optimization; when the mechanical strength error is large, the error compensation factor will automatically increase the optimization weight of the mechanical strength, so that the optimization process pays more attention to the mechanical strength; when the errors of the two are similar, the error compensation factor will be adjusted to a balanced state to make the optimization weights of the two equal.

[0047] Optionally, the error compensation factor can be calculated based on the optimization target weight and the current error ratio, so that it can dynamically adapt to different optimization requirements. During the calculation process, the initial weight set by the optimization target can be referenced, and the weighted average can be combined with the current error size to ensure the rationality of the optimization strategy.

[0048] Furthermore, in traditional neural network optimization, the role of the loss function is to measure the deviation between the predicted value and the true value, and guide the neural network to adjust the weight to reduce the error. To adapt to different optimization modes, the loss function needs to be dynamically adjusted: in the transmittance priority mode, the loss function will increase the weight of the transmittance error, so that the neural network pays more attention to the optimization effect of the transmittance; in the mechanical strength priority mode, the loss function will increase the weight of the mechanical strength error, so that the optimization process is more inclined to improve the mechanical strength; in the balanced optimization mode, the loss function will balance the transmittance error and the mechanical strength error, so that the optimization weights of the two remain relatively stable, which can be expressed as the sum of the transmittance error multiplied by the dynamic weight factor of the transmittance error and the dynamic weight factor of the mechanical strength error multiplied by the mechanical strength error.

[0049] This method can effectively improve the flexibility of the optimization process and enable the neural network to obtain the best optimization effect under different optimization modes.

[0050] In the actual optimization process, the optimization requirements for transmittance and mechanical strength will be dynamically adjusted with changes in material properties, manufacturing processes and environmental factors. If only the initial optimization weights are relied upon for training, it may result in: the optimization strategy being solidified, the optimization direction being unable to be adjusted according to the target requirements, the error weight adjustment being lagging, the optimization direction being unable to be adapted in time, resulting in a slower convergence speed, and the final composite film performance being unable to reach the global optimum due to the bias of a single optimization target, etc.

[0051] Therefore, in the embodiment of the present invention, step S4 adopts two consecutive rounds of optimization iterations. In each round of optimization, not only the error of the optimization target itself is reduced, but also the target weight is dynamically corrected using the error change trend to ensure that the optimization target can be adaptively adjusted according to actual conditions, thereby improving the comprehensive performance of the composite membrane.

[0052] Specifically, before conducting two consecutive rounds of optimization iterations, the following prerequisites need to be ensured: the initial optimization target weight coding has been established, the target weight coding has been calculated through the optimization target contribution, and added to the neural network optimization system as an input feature; the first-layer optimization structure: the neural network input layer adds weight coding to ensure that the optimization target weight can affect the entire optimization process; the second-layer optimization structure: the target sensitivity automatically adjusts the hidden layer activation function to ensure that different optimization targets can adjust the optimization direction under different activation methods; the error compensation factor β is used to control the optimization weights of transmittance and mechanical strength errors, and a loss function that adapts to different optimization modes is established.

[0053] On this basis, if Figure 4 As shown, the first round of optimization iteration: Calculation of the initial error change trend includes: During the first round of optimization, the system needs to perform the following steps: read the initial weight code of the optimization target set by S1; calculate the error compensation factor β, which is used to control the optimized weight ratio of transmittance and mechanical strength; and enter the weight code into the optimization system as part of the input data.

[0054] Perform the first round of optimization training: Use the two-level optimization structure of S2, input data for forward propagation calculation. Calculate the current transmittance error and mechanical strength error; use the loss function set by S3 to update the network weights and complete the error back propagation (BP training).

[0055] Record the error of the current round and calculate the error decrease rate. That is, if the transmittance error decreases faster, it means that the current optimization strategy is effective for transmittance optimization and the transmittance weight can be appropriately reduced. If the mechanical strength error decreases slowly, it means that mechanical strength optimization is more difficult and its optimization weight needs to be increased to strengthen the optimization efforts.

[0056] Calculate the adjusted weight code: If the transmittance error decrease rate is greater than the mechanical strength error decrease rate, then reduce the transmittance error and increase the mechanical strength error to make the optimization process lean towards mechanical strength optimization; if the mechanical strength error decrease rate is greater than the transmittance error decrease rate, then reduce the mechanical strength error and increase the transmittance error to strengthen transmittance optimization; if the two error decrease rates are close: keep the weight unchanged and continue balanced optimization. Calculate the new optimization target weight code and update the input data.

[0057] The second round of optimization iteration: Target adaptive switching includes: First, optimization training is performed based on the adjusted weights: the updated optimization target weights are used for the second round of optimization iterations; during the training process, the error compensation factor β continues to dynamically adjust the optimization strategy to ensure that the error does not overly bias toward a certain optimization target.

[0058] Record the error after the second round of optimization and calculate the error convergence: if the error continues to decrease, it means that the optimization direction is correct and the current weight strategy is maintained; if the error decreases slowly, it means that the optimization may have entered a local optimum and further adjustments to the optimization strategy are required. Set the error convergence threshold. If the error change after two rounds of optimization is less than the set threshold, the optimization is considered to have converged and the weight adjustment is terminated. Otherwise, continue to perform new optimization iterations until the error stabilizes.

[0059] Through the above-mentioned iterative operations, the optimization target can be adaptively adjusted to avoid the solidification of the optimization target, ensuring that the transmittance and mechanical strength of the composite film can reach the optimal state at the same time; improving the optimization accuracy, making the optimization process more accurate through error change trend calculation, reducing invalid training, and improving optimization efficiency; preventing local optimal traps, and avoiding the optimization strategy from converging to a suboptimal solution too early through dynamic adjustment of the target weights; adapting to different optimization needs, whether in transmittance priority mode, mechanical strength priority mode, or balanced optimization mode, the optimization strategy can be automatically adjusted to improve the overall performance of the composite film.

[0060] It should be noted that step S4 adjusts the target weight in combination with the error change trend through two consecutive rounds of optimization iterations, and corrects the weight coding so that the optimization target can be adaptively adjusted according to the optimization requirements. This method can improve the optimization effect of the transmittance and mechanical strength of the composite film, make the final optimization result more stable and reliable, and at the same time improve the optimization efficiency, ensuring that the final optimization direction is correct.

[0061] In summary, the composite film performance parameter optimization method based on the neural network of the embodiment of the present invention is explained, which realizes the coordinated optimization of the transmittance and mechanical strength of the composite film through the optimization target set, the two-stage optimization structure, the dynamic error compensation mechanism and two consecutive rounds of optimization iterations. Compared with the traditional method, the present invention can dynamically adjust the priority of the optimization target according to the optimization requirements, make the optimization process more adaptable, and improve the stability and accuracy of the final optimization result.

[0062] The two-stage optimization structure of the present invention makes the optimization process more targeted. The first layer enables the network to identify different optimization requirements and adjust the optimization strategy accordingly by adding optimization target weight coding to the input layer of the neural network; the second layer adaptively adjusts the hidden layer activation function based on the target sensitivity to improve the adaptability of the neural network to different optimization targets. In addition, the introduction of the dynamic error compensation mechanism ensures that the transmittance and mechanical strength errors are adjusted evenly during the optimization process to avoid excessive bias of a certain optimization target, resulting in a decrease in overall performance. Two consecutive rounds of optimization iterations further improve the adaptability of the optimization strategy, and adjust the target weight coding by adjusting the error change trend, making the optimization process more intelligent and avoiding falling into a local optimal solution. The present invention can achieve higher optimization accuracy and adaptability in optimizing the performance parameters of the composite membrane, and improve the comprehensive performance of the composite membrane.

[0063] In summary, the neural network-based composite membrane performance parameter optimization method based on the embodiment of the present invention is explained, and those skilled in the art will easily think of other embodiments of the present invention after considering the specification and practicing the embodiments disclosed herein. The present invention is intended to cover any variation, use or adaptive change of the present invention, which follows the general principles of the present invention and includes common knowledge or customary technical means in the art that are not disclosed in the present invention.

[0064] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for optimizing composite membrane performance parameters based on neural network, characterized in that: include: Set a set of optimization targets to dynamically adjust the optimization priority of the targets; Construct a two-level optimization structure, the first layer uses the neural network input layer to add weight encoding, and the second layer automatically adjusts the hidden layer activation function based on the target sensitivity to adapt to different optimization requirements; A dynamic error compensation mechanism is used to jointly analyze the target errors and adjust the neural network loss function to control the errors in different optimization modes; Through two consecutive rounds of optimization iterations, the target weight is adjusted, and the weight coding is corrected according to the error variation trend, so as to realize adaptive switching of the optimization target and improve the comprehensive performance of the composite membrane.

2. The method for optimizing composite membrane performance parameters based on neural network according to claim 1, characterized in that: The objectives include light transmittance and mechanical strength.

3. The method for optimizing composite membrane performance parameters based on neural network according to claim 2, characterized in that: The optimization target set is ,in, is the transmittance target, For mechanical strength purposes.

4. The method for optimizing composite membrane performance parameters based on neural network according to claim 1, characterized in that: The optimization priorities of the dynamic control objectives include: Setting optimization priority parameters , controls the weights of transmittance and mechanical strength during the optimization process; Adaptive target weight adjustment based on experimental data; The target weight configuration used in the current optimization process is identified by optimizing the target encoding.

5. The method for optimizing composite membrane performance parameters based on neural network according to claim 1, characterized in that: The two-stage optimization structure includes: First-layer optimization: By adding optimization target weight encoding to the neural network input layer, the network can understand the current optimization priority and distinguish different optimization requirements; Second-layer optimization: Adopting the target sensitivity mechanism, according to the influence of the optimization target on the result, the hidden layer activation function is adaptively adjusted so that the network can adapt to different optimization requirements.

6. The method for optimizing composite membrane performance parameters based on neural network according to claim 5, characterized in that: The second layer optimization includes: The optimization target set G is used to derive the composite membrane structure parameters to express the influence of the changes in different composite membrane structure parameters on the target optimization, that is, the target sensitivity; Establish a target sensitivity matrix based on target sensitivity; According to the calculation results of the target sensitivity matrix, the activation function of the hidden layer is adjusted.

7. The method for optimizing composite membrane performance parameters based on neural network according to claim 1, characterized in that: The dynamic error compensation mechanism is used to jointly analyze the target errors respectively, including: Calculate the transmittance error and mechanical strength error respectively; Ensure the optimization direction is correct according to the error ratio of the transmittance error and the mechanical strength error; The dynamic error compensation factor is calculated to dynamically adjust the optimization weights of transmittance error and mechanical strength error.

8. The method for optimizing composite membrane performance parameters based on neural network according to claim 7, characterized in that: The transmittance error is measured by the difference between the predicted value of the neural network and the actual value measured experimentally; The mechanical strength error is uniformly calculated by comparing the predicted value with the actual value.

9. The method for optimizing composite membrane performance parameters based on neural network according to claim 1, characterized in that: The adjustment of the neural network loss function and the control of the error in different optimization modes include: The optimization modes include a transmittance priority mode, a mechanical strength priority mode and a balanced optimization mode; The loss function is dynamically adjusted in different optimization modes, including: In the transmittance priority mode, the loss function increases the weight of the transmittance error; In the mechanical strength priority mode, the loss function increases the weight of the mechanical strength error; In the balanced optimization mode, the loss function balances the transmittance error and the mechanical strength error so that the optimization weights of the two remain stable.

10. The method for optimizing composite membrane performance parameters based on neural network according to claim 1, characterized in that: The two consecutive rounds of optimization iterations include: In the first round of optimization, the optimization target weight code is read, the error compensation factor β is calculated, and the optimization target weight is dynamically adjusted based on the error change trend to ensure the balance of the optimization strategy; In the second round of optimization, the optimization target weights adjusted in the first round are used for training, and the error compensation factor β is continuously and dynamically adjusted to optimize the error balance and calculate the error convergence. If the error change slows down, the weight is further adjusted, otherwise the optimization convergence is judged to ensure that the final optimization direction is correct.

Images