Performance optimization design method and system for multi-layer composite polarizer

Abstract

The present application relates to the technical field of polarizer performance optimization, and discloses a performance optimization design method and system for a multi-layer composite polarizer, the method comprising: performing polarization spectrum analysis on the use environment of a display screen, obtaining incident polarization spectrum data, and calculating initial stretching ratio data of each polarizing layer; measuring and obtaining molecular orientation data and polarization performance data of each polarizing layer; performing feature extraction and fusion, and obtaining a fusion feature vector of each polarizing layer; calculating the influence of changes in molecular orientation of each polarizing layer on optical properties of different polarization states, and obtaining the polarization state importance weight of each polarizing layer; performing reinforcement learning optimization, and generating a stretching ratio adjustment strategy for each polarizing layer; performing multi-objective optimization and sequence quadratic programming on processing parameters of each polarizing layer, and generating a stretching temperature and speed control sequence for each polarizing layer, thereby realizing precise control of stretching temperature and speed, and improving the performance index and production efficiency of the polarizer.

Description

Technical Field

[0001] The present application relates to the technical field of polarizer performance optimization, and in particular to a performance optimization design method and system for a multi-layer composite polarizer. Background Art

[0002] As a key optical component of display devices, the performance of multi-layer composite polarizers directly affects the clarity, viewing angle and energy efficiency of the displayed image. Traditional polarizer design methods mainly rely on experience accumulation and repeated experiments, which are difficult to adapt to the demand for high-performance polarizers in modern display technology, especially in large-size, high-resolution display devices. The performance optimization of polarizers becomes more complicated and difficult.

[0003] In the current polarizer production process, there are still great challenges in controlling the degree of molecular orientation and optimizing the stretching process parameters. Due to the lack of accurate mathematical models and adaptive control strategies, traditional methods are difficult to accurately predict and control the changes in optical properties under different polarization states, resulting in large fluctuations in product performance and low yield rates, as well as a large amount of material waste and energy consumption. In addition, complex operating environments and variable light source conditions place higher demands on the performance of polarizers. Existing technologies are difficult to effectively balance multiple optimization goals such as polarization efficiency, stress distribution uniformity, and energy consumption, and are unable to achieve precise control and real-time adjustment of process parameters, which seriously restricts the development of high-performance polarizers and the improvement of production efficiency. Summary of the invention

[0004] The present application provides a performance optimization design method and system for a multi-layer composite polarizer, thereby achieving precise control of stretching temperature and speed, and improving the performance indicators and production efficiency of the polarizer.

[0005] The first aspect of the present application provides a performance optimization design method for a multi-layer composite polarizer, and the performance optimization design method for a multi-layer composite polarizer includes:

[0006] Perform polarization spectrum analysis on the use environment of the display screen, obtain incident polarization spectrum data, and calculate the initial stretch ratio data of each polarizing layer;

[0007] The production process data of each polarizing layer of the sample product is measured respectively to obtain the molecular orientation data and polarization performance data of each polarizing layer;

[0008] The molecular orientation data and polarization performance data of each polarizing layer are respectively input into the dual-domain and global context feature extraction networks for feature extraction and fusion, and the fused feature vector of each polarizing layer is obtained;

[0009] Based on the fused feature vector, the influence of the change of the molecular orientation degree of each polarizing layer on the optical properties of different polarization states is calculated to obtain the importance weight of the polarization state of each polarizing layer;

[0010] According to the polarization state importance weight, the initial stretch ratio data is optimized by reinforcement learning to generate a stretch ratio adjustment strategy for each polarizing layer;

[0011] Based on the stretching ratio adjustment strategy, multi-objective optimization and sequence quadratic programming are performed on the processing parameters of each polarizing layer to generate a stretching temperature and speed control sequence for each polarizing layer.

[0012] A second aspect of the present application provides a performance optimization design system for a multi-layer composite polarizer, the performance optimization design system for a multi-layer composite polarizer comprising:

[0013] An analysis module is used to perform polarization spectrum analysis on the use environment of the display screen, obtain incident polarization spectrum data, and calculate the initial stretch ratio data of each polarizing layer;

[0014] The measurement module is used to measure the production process data of each polarizing layer of the sample product to obtain the molecular orientation data and polarization performance data of each polarizing layer;

[0015] A fusion module, used to input the molecular orientation data and polarization performance data of each polarizing layer into the dual-domain and global context feature extraction networks for feature extraction and fusion, and obtain a fused feature vector of each polarizing layer;

[0016] A calculation module, for calculating the influence of the change of the molecular orientation degree of each polarizing layer on the optical properties of different polarization states based on the fused feature vector, and obtaining the importance weight of the polarization state of each polarizing layer;

[0017] An optimization module, configured to perform reinforcement learning optimization on the initial stretch ratio data according to the polarization state importance weight, and generate a stretch ratio adjustment strategy for each polarizing layer;

[0018] A generation module is used to perform multi-objective optimization and sequence quadratic programming on the processing parameters of each polarizing layer based on the stretching ratio adjustment strategy, and generate a stretching temperature and speed control sequence for each polarizing layer.

[0019] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned performance optimization design method of the multi-layer composite polarizer.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned performance optimization design method for the multi-layer composite polarizer.

[0021] Compared with the prior art, the present application has the following beneficial effects: by introducing dual-domain and global context feature extraction networks for feature extraction and fusion, combined with reinforcement learning optimization and multi-objective sequential quadratic programming, precise optimization of polarizer performance and adaptive control of process parameters are achieved. This method has three outstanding technical advantages: through dual-path feature extraction in the spatial domain and frequency domain, combined with global context feature encoding, comprehensive capture of molecular orientation and polarization performance characteristics is achieved; a reinforcement learning method based on polarization state importance weights is adopted to establish an adaptive optimization mechanism for the stretching ratio parameters, making the adjustment of process parameters more accurate; through multi-objective optimization and sequential quadratic programming, combined with segmented control and transition optimization strategies, precise control of stretching temperature and speed is achieved, and the robustness of the process is enhanced through the main-standby control sequence switching mechanism. The present invention significantly improves the performance indicators and production efficiency of polarizers. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0024] Figure 1 It is a schematic flow chart of a performance optimization design method for a multi-layer composite polarizer provided by an embodiment of the present invention;

[0025] Figure 2 is a schematic structural block diagram of a performance optimization design system for a multi-layer composite polarizer provided by an embodiment of the present invention;

[0026] Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0029] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0030] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the performance optimization design method of the multilayer composite polarizer in the embodiment of the present application includes:

[0031] Step 100, performing polarization spectrum analysis on the use environment of the display screen, obtaining incident polarization spectrum data, and calculating initial stretch ratio data of each polarizing layer;

[0032] It is understandable that the execution subject of the present application can be a performance optimization design system for a multi-layer composite polarizer, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0033] Specifically, the natural light source in the display screen's use environment is measured at multiple angles to obtain spatial distribution data of the incident light intensity. The spatial distribution information of light intensity is the basis for understanding the interaction between the light source and the display screen. The spatial distribution data of the incident light intensity is decomposed into polarization states, and the horizontal polarization component and vertical polarization component data of the light wave are extracted from them. Spectral analysis is performed based on the horizontal and vertical polarization component data to obtain the distribution data of each polarization state at different wavelengths. By analyzing these data, the wavelength distribution characteristics of the polarization state are obtained. An optimization objective function for polarized light transmittance is constructed based on the wavelength distribution data. The optimization objective function maximizes the transmittance of polarized light by adjusting the design parameters of the polarizer, and the polarized light optimization weight matrix is ​​calculated based on this objective function. The weight matrix reflects the relative importance of different polarization states to the optimization of light transmittance. The optical performance transfer equation of the multilayer composite polarizer is constructed based on the polarized light optimization weight matrix. The equation describes the relationship between the polarization efficiency and optical performance of each polarizing layer, and the polarization efficiency target value of each polarizing layer is calculated through the transfer equation. The target value is obtained based on the optical design requirements and actual production conditions for the performance requirements of each polarizing layer. In order to ensure that the design target of each polarizing layer can be achieved in actual production, stress-strain analysis is performed on the polarization efficiency target value. Through stress-strain analysis, the stress conditions of each polarizing layer in the production process are simulated, and the molecular orientation data required for each polarizing layer is obtained. Based on the molecular orientation data, the mechanical properties of the material are analyzed to evaluate the physical properties of each layer of the polarizer under different stresses, including the elastic modulus, yield strength, plastic deformation characteristics, etc. of the material. By analyzing the stress distribution data of each polarizing layer, the stress data that each polarizing layer needs to withstand in the actual production process is obtained. Through these stress data, the process parameters such as the stretching temperature, speed, and stretching ratio of each polarizing layer are calculated, which directly determine the performance and quality of the final polarizer. Combining stress analysis and material mechanical properties, the initial stretching ratio data of each polarizing layer is calculated to ensure that the best molecular orientation and polarization performance can be obtained during the production process.

[0034] Step 200, measuring the production process data of each polarizing layer of the sample product respectively to obtain the molecular orientation data and polarization performance data of each polarizing layer;

[0035] Specifically, the transmittance is measured by a multi-point polarization spectrum analyzer, and the scanning wavelength range is set to 450-750nm. The wavelength range covers the main part of visible light and can effectively capture the optical performance of the polarizer at different wavelengths. The transmittance of the sample product is measured by the device to obtain the light intensity distribution data of each polarizing layer at different wavelengths, reflecting the change in light intensity when the light passes through each polarizing layer. Polarization state separation calculation is performed based on the light intensity distribution data. The transmittance data of different polarization states are extracted to obtain parallel transmittance and vertical transmittance data. Parallel transmittance and vertical transmittance reflect the efficiency of polarized light transmission along different directions when light passes through the polarizer. Through polarization state separation calculation, the transmittance characteristics of the polarizer in different polarization states are obtained. The parallel transmittance data and the vertical transmittance data are input into the extinction ratio calculation model. The extinction ratio is an important parameter of the performance of the polarizer, which measures the difference in transmittance of the polarizer in the parallel polarization and vertical polarization states of light. The calculated polarization performance data can intuitively reflect the extinction performance of the polarizer. The sample product is scanned at the incident angle of the ellipsometer to obtain the amplitude ratio data and phase difference data of each polarizing layer. These data are important parameters for describing the propagation characteristics of light in the polarizer material. The amplitude ratio reflects the intensity change of light, while the phase difference indicates the phase delay of different polarization components when passing through the material. Based on the amplitude ratio data and phase difference data, the birefringence calculation is performed to obtain the refractive index anisotropy data of each polarizing layer, which describes the refractive index difference of the material under different polarization states, thereby affecting the propagation and polarization effect of light. The molecular orientation degree is inverted and calculated for the refractive index anisotropy data, and the molecular orientation degree of the material is inferred from the refractive index anisotropy data. The molecular orientation degree reflects the arrangement state of the material molecules during the production process and is one of the key factors affecting the performance of the polarizer. The molecular orientation degree data of each polarizing layer is obtained through inversion calculation. According to the molecular orientation degree data, the polarization performance of each polarizing layer is theoretically simulated to predict its performance under different optical conditions and obtain theoretical polarization performance data. Compare and analyze the theoretical polarization performance data with the polarization performance data obtained through experimental measurement to evaluate the degree of consistency between the polarization performance of the actual sample and the theoretical model, thereby verifying the accuracy of the molecular orientation data. If there is a deviation between the theoretical polarization performance data and the experimental data, the performance of the polarizer can be improved by adjusting the molecular orientation data and optimizing the design parameters.

[0036] Step 300: input the molecular orientation data and polarization performance data of each polarizing layer into the dual-domain and global context feature extraction networks for feature extraction and fusion, and obtain a fused feature vector of each polarizing layer;

[0037] It should be noted that the molecular orientation data of each polarization layer is input into the spatial domain feature extraction branch. This branch contains four convolutional layers, each of which uses a convolution operation with a step size of 2 and is normalized in combination with BatchNorm to ensure that the features of each layer remain stable during training. The convolutional layer uses LeakyReLU as the activation function, which can effectively avoid the gradient vanishing problem and enhance the network's nonlinear expression ability of features. Through this series of operations, the spatial domain feature extraction branch generates a spatial domain feature map that reflects the detailed information of the molecular orientation in the spatial domain. The molecular orientation data is fast Fourier transformed to extract more information from the frequency domain perspective. Through Fourier transform, the frequency domain representation of the data can reveal the periodicity and frequency components hidden in the original signal. The converted frequency domain data is input into the frequency domain feature extraction branch, which contains four convolutional layers. The frequency domain feature extraction branch uses the dilated convolution technique to enhance the receptive field of the convolution operation, and at the same time, through the SE attention mechanism, it improves the network's ability to capture important information in the frequency domain feature extraction process. The SE attention mechanism can automatically learn the relationship between channels, dynamically adjust the weights of each channel, strengthen meaningful features, and suppress useless information. The frequency domain feature extraction branch generates frequency domain feature maps to capture important information in the frequency domain. The spatial domain and frequency domain feature maps are weighted through the channel attention weighting mechanism. The spatial domain and frequency domain features are fused using the cross-domain feature fusion module. The information from two different domains is combined to generate a dual-domain feature vector, which fully integrates the advantages of the spatial and frequency domains and enhances the expressiveness of the model. The polarization performance data is input into the global context encoder. The encoder contains three multi-head self-attention layers, each of which consists of eight attention heads. The multi-head self-attention mechanism can perform parallel calculations between different attention heads, thereby capturing different context information in different subspaces and enhancing the learning ability of the model. Through this mechanism, the model can capture long-range dependencies in the polarization performance data on a global scale and generate a global context feature vector. The global context feature vector is subjected to multi-scale feature decomposition. The decomposition is performed through three parallel 1×1 convolutional layers, each of which extracts context information at different scales. 1×1 convolution has the characteristics of few parameters and high computational efficiency. It effectively extracts information from feature maps of different scales, thereby improving the model's sensitivity to multi-scale contexts and obtaining multi-scale context features. The dual-domain feature vector and multi-scale context features are input into the feature fusion network, which contains two fully connected layers and a residual connection. The residual connection effectively prevents the gradient vanishing problem and facilitates information transfer during model training, making the fusion process more stable and efficient. Through this feature fusion network, the initial fusion features are generated. The initial fusion features are self-calibrated. The long-range dependencies are captured through the non-local attention module to optimize the fusion features.The non-local attention module can interact information between various positions in the feature map, capture the relationship between distant pixels, and enhance the expression of global information. The calibrated fused features are input into the feature compression network, which consists of two fully connected layers and a Dropout layer to compress the features and prevent overfitting, and obtain the fused feature vector of each polarization layer.

[0038] Step 400: Based on the fused feature vector, calculate the influence of the change of the molecular orientation of each polarizing layer on the optical properties of different polarization states, and obtain the importance weight of the polarization state of each polarizing layer;

[0039] Specifically, singular value decomposition is performed on the fused eigenvector, and the eigenvector is decomposed into several eigencomponents. The contribution rate of each eigencomponent is determined by singular value decomposition. The eigencomponents with cumulative contribution rates greater than 95% are selected, and the most important information is retained. These components constitute the eigencomponent matrix of each polarizing layer. The eigencomponent matrix is ​​standardized, and the data is normalized to the same range to obtain standardized eigendata. Based on the standardized eigendata, a molecular orientation change sequence is constructed. The molecular orientation of each polarizing layer is increased by 1°, 2°, 3°, and decreased by 1°, 2°, and 3° on the basis of the baseline value, respectively, to obtain 7 sets of molecular orientation change data. These data represent the changes in the molecular orientation of the polarizing layer under different adjustment ranges, providing a range of changes for subsequent calculations. The 7 sets of molecular orientation change data are substituted into the birefringence calculation formula. The formula is: n=n0+Δn(3cos2θ-1) / 2, where n is the birefringence, n0 is the isotropic refractive index, Δn is the maximum birefringence, and θ is the molecular orientation angle. By substituting different molecular orientation change data, the birefringence data of each group under different molecular orientation degrees are calculated. Jones matrix calculation is performed for each group of birefringence data. Jones matrix is ​​a tool to describe how light waves change their polarization states after passing through various optical media. In the calculation process, it is assumed that the incident light is in four polarization states of 0°, 45°, 90°, and 135°. The transmittance values ​​of each group of birefringence data in these four polarization states are calculated respectively to obtain the corresponding relationship data between polarization state and transmittance. Through this process, the transmittance change of the polarizer in different polarization states under different molecular orientation changes is obtained. Based on the polarization state-transmittance correspondence data, the transmittance change caused by the change of each group of molecular orientation is calculated. The sensitivity of transmittance to the change of molecular orientation is calculated by the central difference method, that is, by performing differential operations on the data of different orientation changes, the change rate of transmittance is obtained, thereby constructing an optical performance response curve. This curve represents the response degree of the optical performance of the polarizer to the change of molecular orientation. The optical performance response curve is integrated. In the range of 0° to 180°, the area under the curve is calculated at intervals of 5° to obtain the cumulative impact value of the optical performance of each polarization state, reflecting the overall impact of changes in molecular orientation on the optical performance of different polarization states. The cumulative impact value of the optical performance is normalized, and the maximum value normalization is used. All cumulative impact values ​​are standardized according to the maximum value, so that the impact values ​​between different polarizing layers can be compared on the same scale. The standardized feature data is matrix multiplied with the normalized cumulative impact value of the optical performance to obtain the initial weight data of the polarization state, indicating the performance importance of each polarizing layer in different polarization states. The exponential function exp(x) is used to numerically map the initial weight to an exponential range, thereby enhancing the expression of importance.At the same time, in order to avoid numerical instability, L2 norm normalization is used to standardize the weight data so that it finally meets the predetermined scale and range. The obtained polarization state importance weight is the performance weight of each polarization layer in different polarization states.

[0040] Step 500: Perform reinforcement learning optimization on the initial stretch ratio data according to the polarization state importance weight to generate a stretch ratio adjustment strategy for each polarizing layer;

[0041] Specifically, a state space matrix is ​​constructed according to the polarization state importance weight of each polarization layer and the initial stretch ratio data. The state vector of the state space matrix contains three components: the current stretch ratio, the target stretch ratio, and the polarization state importance weight. Through these data, the initial state data is obtained. The initial state data is input into the policy network. The network consists of three fully connected layers, of which the first and second layers use the ReLU activation function to increase the nonlinear expression ability of the network and can capture complex relationships; while the third layer uses the tanh activation function to limit the range of the action and constrain it within an acceptable range to ensure that the generated stretch ratio adjustment action does not exceed the predetermined operating range. Through the operation of the policy network, the stretch ratio adjustment action data is obtained, which represents the optimal stretch ratio adjustment strategy that should be taken in the current state. The generated stretch ratio adjustment action data is discretized. The continuous stretch ratio adjustment space is divided into 21 discrete action points, which cover the range of stretch ratio adjustment, with a step size of 0.1 and an adjustment range of [-1, 1]. Discretization converts the continuous action space into finite discrete actions, thereby simplifying the decision-making problem in reinforcement learning and making each action clearly correspond to a certain adjustment value of the stretch ratio. The discretized action data is input into the value evaluation network, which contains four fully connected layers. Each layer uses the LeakyReLU activation function to improve the stability of the gradient during training and avoid the gradient vanishing problem. The last layer outputs the Q value, which represents the expected value of each action in the current state, and obtains the value data of the action. The action value data is decomposed into a double advantage function, and the state value function and the action advantage function are processed separately. The state value function represents the overall value in the current state, while the action advantage function reflects the advantage of performing an action over other actions in a specific state. In this process, the attention mechanism is used to weight the importance of different actions, highlight the actions that have a greater contribution to performance optimization, and obtain advantage weighted data. Based on the advantage weighted data, a priority experience replay buffer is constructed. The function of this buffer is to store the experience data obtained by the agent during execution and assign sampling weights to each action according to its importance score. In this way, high-value experience data is sampled first to avoid inefficient training processes. During the sampling process, a stratified sampling method is used to select the empirical trajectory that has the greatest impact on performance improvement and obtain the optimized trajectory data. The optimized trajectory data is input into the target network. The structure of the target network is the same as that of the policy network. The network parameters are updated by soft update, and the update rate is set to tau=0.01. Soft update can avoid large-scale updates of network weights and ensure the stability of the training process, thereby gradually improving the strategy. Based on the difference between the target action data and the discretized action data, the temporal difference error is calculated. The temporal difference error is an important indicator in reinforcement learning, which is used to measure the deviation between the current strategy and the target strategy.In order to ensure that the optimization process can converge effectively, the timing differential error and the polarization state importance weight are weighted averaged to make a more reasonable balance in the optimization effects of different polarization states. This ensures that during the optimization process, the polarization state that has a greater impact on the optical performance is focused on, generating a more accurate and effective stretch ratio adjustment strategy.

[0042] Step 600: Perform multi-objective optimization and sequence quadratic programming on the processing parameters of each polarizing layer based on the stretching ratio adjustment strategy to generate a stretching temperature and speed control sequence for each polarizing layer.

[0043] Specifically, two sets of process constraint matrices are constructed based on the stretching ratio adjustment strategy. The first process constraint matrix covers the stretching temperature range and the heating rate range, where the stretching temperature is controlled between [120℃, 180℃] and the heating rate is controlled between [1℃ / min, 5℃ / min] to ensure proper temperature control and avoid instability of polarizer performance due to too high or too low temperature. The second process constraint matrix contains the stretching speed range [5mm / min, 50mm / min] and the stretching acceleration range [0.1mm / min², 1mm / min²]. The constraints are to control the deformation rate and acceleration of the polarizer during the stretching process to ensure stability and accuracy during the processing. These constraints are used as optimization boundary conditions to help determine the feasible processing range. The first constraint and the second constraint are constructed for collaborative optimization objectives. The first optimization objective is set to maximize polarization efficiency. The second optimization objective is set to maximize the uniformity of stress distribution. During the stretching process, if the stress distribution of the polarizer is too concentrated, it will lead to instability or even rupture of the optical performance, so it is necessary to ensure the uniformity of stress distribution. The third optimization goal is to minimize energy consumption. Energy consumption in the stretching process is directly related to production costs. By optimizing parameters such as heating rate and temperature, minimizing energy consumption is optimization. By combining these three optimization goals, a comprehensive multi-objective optimization function is obtained. Based on the multi-objective optimization function, two control variable matrices are constructed. The first control variable matrix describes the relationship between temperature and time to ensure that the stretching process can proceed stably within the preset temperature range; the second control variable matrix describes the relationship between speed and displacement to ensure the synchronization of speed control and displacement during the stretching process to achieve precise control. These two matrices together constitute the global optimization variable space, providing global optimization direction and goals. Segmented optimization is solved based on the global optimization variable space. The stretching process is divided into three main stages: preheating stage, stretching stage, and cooling stage. The optimal control parameters of each stage are calculated according to the actual processing requirements to obtain the optimal control parameter sequence for each stage. For example, in the preheating stage, the main optimization is the control of heating rate and temperature; in the stretching stage, the focus is on optimizing the stretching speed and stress distribution; in the cooling stage, the cooling rate needs to be optimized to ensure that after the stretching process, the material can cool smoothly without stress concentration. By optimizing each segment, a segmented control parameter sequence is obtained. The transition segments of the adjacent segments of the segmented control parameter sequence are optimized to avoid performance degradation or material damage in the stretching process due to sudden changes. The transition function is introduced to control the continuity of temperature and speed. The first transition function is used to control the continuity of temperature change to avoid sudden temperature changes between different process segments; the second transition function is used to control the continuity of speed change to ensure that the stretching speed change in the transition segment is smooth to avoid adverse effects on the performance of the polarizer.The first parameter compensation model and the second parameter compensation model are constructed based on the smooth transition sequence. The first parameter compensation model is used to correct the temperature hysteresis. There is usually a hysteresis in the temperature change during the stretching process. The temperature control is compensated to ensure the match between the actual temperature and the target temperature. The second parameter compensation model is used to correct the speed response. Because the actual response of the stretching speed usually has a certain delay, the speed is corrected to ensure that the speed change is in line with expectations. The compensation optimization parameters are converted into control instructions according to the first quantization accuracy and the second quantization accuracy. The minimum division value of temperature control is 0.1℃, and the minimum division value of speed control is 0.1mm / min, and the digital control instructions are obtained. The timing combination is arranged according to the digital control instructions to generate a dual-sequence control scheme including the main control sequence and the backup control sequence. The main control sequence is suitable for standard working conditions and can ensure the optimal stretching process under normal production conditions; the backup control sequence is used for abnormal working condition switching. For example, when the equipment fails or the temperature changes greatly, it can respond quickly and switch to the backup scheme to ensure the continuity and safety of the production process. The dual-sequence control scheme ensures that appropriate stretching temperature and speed control sequences can be obtained under different working conditions, ensuring that the stretching process of each polarizing layer can be carried out under optimal conditions, ultimately achieving the optimization of the performance of the multi-layer composite polarizer and obtaining the stretching temperature and speed control sequence of each polarizing layer.

[0044] In the embodiment of the present application, by introducing dual-domain and global context feature extraction networks for feature extraction and fusion, combined with reinforcement learning optimization and multi-objective sequential quadratic programming, precise optimization of polarizer performance and adaptive control of process parameters are achieved. This method has three outstanding technical advantages: through dual-path feature extraction in the spatial domain and frequency domain, combined with global context feature encoding, comprehensive capture of molecular orientation and polarization performance characteristics is achieved; a reinforcement learning method based on polarization state importance weights is adopted to establish an adaptive optimization mechanism for the stretching ratio parameters, making the adjustment of process parameters more accurate; through multi-objective optimization and sequential quadratic programming, combined with segmented control and transition optimization strategies, precise control of stretching temperature and speed is achieved, and the robustness of the process is enhanced through the main-standby control sequence switching mechanism. The present invention significantly improves the performance indicators and production efficiency of polarizers.

[0045] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0046] Perform multi-angle light intensity measurement on the natural light source in the use environment of the display screen to obtain spatial distribution data of incident light intensity, and perform polarization state decomposition on the spatial distribution data of incident light intensity to obtain horizontal polarization component data and vertical polarization component data;

[0047] Perform spectral analysis on the horizontal polarization component data and the vertical polarization component data to obtain the wavelength distribution data of each polarization state, and construct a polarization light transmittance optimization objective function based on the wavelength distribution data to obtain a polarization light optimization weight matrix;

[0048] The optical performance transfer equation of the multi-layer composite polarizer is constructed according to the polarization light optimization weight matrix to obtain the polarization efficiency target value of each polarizing layer, and the stress-strain analysis of the polarization efficiency target value is performed to obtain the molecular orientation data required for each polarizing layer;

[0049] The mechanical properties of the material are analyzed based on the molecular orientation data to obtain the stress distribution data of each polarizing layer, and the stretching process parameters are calculated based on the stress distribution data to obtain the initial stretching ratio data of each polarizing layer.

[0050] Specifically, a multi-angle light intensity measuring instrument is used to measure the natural light source in the display screen usage environment. In this measurement, the light intensity value at each incident angle is Measured by the light intensity sensor, represents the polar angle of the incident light, represents the azimuth of the incident light, It represents the light intensity at that angle. These data are recorded by multi-point measurement to obtain the spatial distribution data of the light intensity of the entire light source. The spatial distribution data of the light intensity of the incident light is decomposed by polarization state. The different polarization components in the natural light source are separated into horizontal polarization components and vertical polarization components. Polarization information is extracted from the incident light intensity data through Fourier transform or spectral decomposition technology. It is expressed as the superposition of horizontal and vertical polarization components:

[0051] ;

[0052] in represents the horizontal polarization component, represents the vertical polarization component. In this way, the data of the horizontal polarization component and the vertical polarization component are obtained respectively. and the vertical polarization component Perform spectral analysis to obtain the distribution of different polarization states at different wavelengths. Scan the wavelength of the horizontal and vertical polarization components with a spectrometer to obtain the wavelength distribution data of each polarization state. Set the wavelength range to (e.g., 400nm to 700nm), for each polarization state and , and obtain the corresponding wavelength distribution data:

[0053] ;

[0054] ;

[0055] in, is the spatial angle element, Represents wavelength. Through spectral analysis, the transmittance distribution of each polarization state at different wavelengths is obtained. Based on the wavelength distribution data, the optimization objective function of polarized light transmittance is constructed. This objective function aims to maximize the transmittance of each polarization state, while optimizing the polarization performance of different wavelengths to ensure the good performance of the polarizer at different wavelengths. The optimization objective function is expressed as:

[0056] ;

[0057] in, are weighting factors associated with different wavelengths, is at wavelength By optimizing this function, we can get the polarized light optimization weight matrix: , which contains the optimized weights for different wavelengths and polarization states. , construct the optical performance transfer equation of the multi-layer composite polarizer. This equation is used to describe the polarization effect of light passing through each polarizing layer. The optical performance transfer equation of the polarizer is expressed as:

[0058] ;

[0059] in, is the polarization state of the output light, is the polarization efficiency of each polarizing layer, is the polarization optimization weight. By constructing this transfer equation, the polarization efficiency target value of each polarizing layer is obtained. The polarization efficiency target value is the expected value of the polarization transmission efficiency that each polarizing layer should achieve under specific conditions, which is obtained through numerical simulation or experimental measurement. Stress-strain analysis is performed on the polarization efficiency target value to ensure that the polarizing layer will not be excessively deformed or stressed during the stretching process, thereby affecting its optical performance. The stress-strain analysis uses the finite element analysis method to simulate the stress distribution of the polarizing layer during the stretching process. Assume that the stress distribution of the polarizing layer is represented by a two-dimensional or three-dimensional stress field as ,in are spatial coordinates, and is the stress value. Based on these data, the molecular orientation data required for each polarizing layer is calculated, that is, the optimal arrangement of the molecular direction of the polarizing layer. The molecular orientation is calculated by the following formula:

[0060] ;

[0061] Determine the optimal molecular orientation angle by maximizing the uniform distribution of stress , so that the polarizer can maintain high polarization efficiency while avoiding damage caused by stress concentration during the stretching process. Based on the molecular orientation data, the mechanical properties of the material are analyzed to obtain the stress distribution data of each polarizing layer, and the stretching process parameters are calculated based on these data. The calculation of the stretching process parameters involves multiple factors such as temperature, speed, acceleration, etc. The stretching ratio calculation formula is:

[0062] ;

[0063] in, is the stretching ratio, and They represent the length of the polarizer after stretching and the initial length, respectively. Based on these calculation results, the initial stretching ratio data of each polarizing layer is obtained.

[0064] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0065] The multi-point polarization spectrum analyzer is set to a wavelength scanning range of 450-750nm, and the transmittance of the sample product is measured to obtain the light intensity distribution data of each polarizing layer at different wavelengths;

[0066] Polarization state separation calculation is performed based on the light intensity distribution data to obtain parallel transmittance data and vertical transmittance data;

[0067] The parallel transmittance data and the perpendicular transmittance data are input into the extinction ratio calculation model to obtain the polarization performance data of each polarizing layer;

[0068] Perform an ellipsometer incident angle scan on the sample product to obtain the amplitude ratio data and phase difference data of each polarizing layer, and perform birefringence calculation based on the amplitude ratio data and phase difference data to obtain the refractive index anisotropy data of each polarizing layer;

[0069] Perform molecular orientation inversion calculation on the refractive index anisotropy data, and simulate the polarization performance of each polarizing layer based on the molecular orientation data to obtain theoretical polarization performance data;

[0070] The theoretical polarization performance data is compared and analyzed with the polarization performance data to determine the molecular orientation data of each polarizing layer.

[0071] Specifically, a multi-point polarization spectrum analyzer is used to measure the transmittance of the sample product. The wavelength scanning range of the analyzer is set between 450 and 750nm, covering the spectrum of the visible light region. Through multi-point measurement, the light intensity distribution data of the sample at different wavelengths is obtained. The set measurement conditions include multiple different incident angles and polarization states. At each measurement point, the light intensity Used to describe the intensity of light transmitted, especially at different wavelengths ( The light intensity under the incident light is obtained by using these data to obtain the light intensity distribution of each polarizing layer at different wavelengths. Polarization state separation calculation is performed based on the light intensity distribution data to extract the horizontal polarization and vertical polarization components from the incident light intensity. Assuming that the transmittance is at different wavelengths The distribution under and , respectively represent the transmittance of the horizontal polarization component and the vertical polarization component. Polarization state separation is achieved through the following relationship:

[0072] ;

[0073] ;

[0074] in, and are the light intensities of the horizontal and vertical polarization components, respectively, is the total light intensity. By calculating these components, the parallel transmittance at different wavelengths is obtained and vertical transmittance The parallel transmittance data and the vertical transmittance data are input into the extinction ratio calculation model to calculate the polarization performance of the polarizer. It is an important parameter to measure the effect of polarizer, which is defined as the ratio of parallel transmittance to vertical transmittance, that is:

[0075] ;

[0076] This ratio can reflect the transmittance of different polarization states at specific wavelengths for different polarization layers. By calculating the extinction ratio, polarization performance data is provided for each polarization layer. The higher the extinction ratio of the polarizer, the stronger its ability to block unwanted polarization states, thus having a better polarization effect. Perform an ellipsometer incident angle scan on the sample product. The ellipsometer can measure the amplitude ratio and phase difference data at different incident angles to obtain the birefringence data of the polarizer. When performing an ellipsometer scan, the amplitude ratio measured is and phase difference As the basis for calculating birefringence. It can be calculated by the following formula:

[0077] ;

[0078] in, is the angle of incidence The amplitude ratio under is the phase difference. The anisotropy of the refractive index is calculated through the birefringence data. Refractive index anisotropy refers to the difference in the refractive index of a material in different directions. For multi-layer polarizers, the anisotropy of the material directly affects the quality of the polarization performance. Using the known refractive index data, the molecular orientation data required for each polarizing layer is obtained through molecular orientation inversion calculation. Molecular orientation is a measure that describes the direction of molecular arrangement of a material, and is derived from the birefringence data through an inversion algorithm. Molecular orientation It is obtained by the following inversion calculation method:

[0079] ;

[0080] in, and are the refractive indices in the horizontal and vertical directions, is the difference in refractive index. The ideal molecular orientation of each polarizing layer is obtained through calculation. Polarization performance simulation is performed based on the molecular orientation data. Through numerical simulation or experimental measurement, the optical performance parameters such as polarization transmittance and extinction ratio of each polarizing layer are calculated based on the change in molecular orientation. The simulation results reflect the impact of different molecular orientations on the overall polarization performance of the polarizer. The simulation obtains theoretical polarization performance data, which represents the maximum polarization efficiency and minimum extinction ratio that the polarizer can achieve under a given molecular orientation. The theoretical polarization performance data is compared and analyzed with the actual polarization performance data measured by experiment, the difference between the experimental data and the theoretical prediction is evaluated, and the parameters in the model are optimized.

[0081] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0082] The molecular orientation data is input into the spatial domain feature extraction branch. The spatial domain feature extraction branch contains 4 convolutional layers. Each convolutional layer uses a convolution operation with a step size of 2 and BatchNorm normalization, and uses the LeakyReLU activation function to obtain a spatial domain feature map.

[0083] The molecular orientation data is fast Fourier transformed and input into the frequency domain feature extraction branch. The frequency domain feature extraction branch contains 4 convolutional layers. Each convolutional layer uses dilated convolution and SE attention mechanism to obtain the frequency domain feature map.

[0084] Perform channel attention weighting on the spatial domain feature map and the frequency domain feature map, and perform feature fusion through the cross-domain feature fusion module to obtain a dual-domain feature vector;

[0085] The polarization performance data is input into the global context encoder, which contains three multi-head self-attention layers, each of which contains 8 attention heads, to obtain a global context feature vector.

[0086] Perform multi-scale feature decomposition on the global context feature vector, extract context information of different scales through three parallel 1×1 convolutional layers, and obtain multi-scale context features;

[0087] The dual-domain feature vector and multi-scale context features are input into the feature fusion network, which contains 2 fully connected layers and 1 residual connection to obtain the initial fusion features;

[0088] The initial fused features are self-calibrated, and the long-range dependencies are captured through the non-local attention module to obtain the calibrated fused features. The calibrated fused features are input into the feature compression network, and the fused feature vector of each polarization layer is obtained through two fully connected layers and a Dropout layer.

[0089] Specifically, the molecular orientation data is taken as input and passed to the spatial domain feature extraction branch. In this branch, 4 convolutional layers are used, and the convolution operation step size of each convolutional layer is set to 2 to increase the network's receptive field and enhance its ability to extract spatial features. BatchNorm normalization is performed after each convolutional layer to reduce internal covariate shift and improve the stability of network training. LeakyReLU activation function is used as the nonlinear activation function to effectively avoid the gradient vanishing problem encountered by traditional ReLU during training. After this step, the spatial domain feature map is obtained to capture the spatial distribution characteristics of the molecular orientation data. The molecular orientation data is fast Fourier transformed to convert it from the spatial domain to the frequency domain. Through the frequency domain feature extraction branch, the periodicity and local features in the data are extracted from the frequency perspective. This branch contains 4 convolutional layers, each of which uses a dilated convolution to expand the receptive field of the convolution kernel without increasing the amount of calculation. The frequency domain feature extraction branch combines the Squeeze-and-Excitation (SE) attention mechanism. The SE mechanism adaptively adjusts the weight of each channel by learning the dependencies between channels, strengthens the focus on important frequency features, and suppresses unimportant features. The spatial domain feature map and the frequency domain feature map are weighted by channel attention, and the channels of each feature map are weighted by the attention mechanism to strengthen the response to important features. The spatial domain and frequency domain feature maps are fused using the cross-domain feature fusion module. This module can obtain a richer dual-domain feature vector by combining the features of two different domains, containing all key information at the spatial and frequency levels. The polarization performance data is input into the global context encoder. The encoder contains three multi-head self-attention layers, each with eight independent attention heads. The multi-head self-attention mechanism can extract information from different subspaces and merge them to capture the global dependencies of the input data by calculating multiple attention heads in parallel. The self-attention mechanism can dynamically learn the correlation between each feature and effectively transfer and aggregate information to better express the global context information. Through this encoder, a global context feature vector is obtained, which reflects the long-range dependencies and global features in the data. The global context feature vector is subjected to multi-scale feature decomposition. Through three parallel 1×1 convolutional layers, contextual information of different scales is extracted respectively. Capturing contextual features from multiple scales helps the network better understand the changes and patterns of data at different scales. Through multi-scale contextual information, the network can more flexibly handle the performance of different polarizing layers under different working conditions, and improve the prediction accuracy of polarizer performance. The dual-domain feature vector and the multi-scale context feature vector are input into the feature fusion network. The network contains two fully connected layers and one residual connection. The function of the fully connected layer is to linearly transform the input feature vector and process it through a nonlinear activation function to generate a more discriminative feature representation.Residual connections can alleviate the gradient vanishing problem during training and help the network learn more stable and effective feature representations. After processing, the initial fused features are obtained. The initial fused features are self-calibrated. The self-calibration operation captures long-range dependencies through the non-local attention module to better adjust the weight distribution of features. The non-local attention module can consider feature interactions in a global range and dynamically adjust the correlation between different features. The calibrated fused features are input into the feature compression network. The feature compression network contains 2 fully connected layers and a Dropout layer. The Dropout layer randomly discards the output of a part of the neurons during training to avoid overfitting and improve the generalization ability of the model. The feature compression network compresses high-dimensional features into low-dimensional representations, extracts the most representative feature vectors, and finally obtains the fused feature vectors of each polarization layer.

[0090] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0091] Perform singular value decomposition on the fused feature vector, select feature components with cumulative contribution rates greater than 95%, generate a feature component matrix for each polarization layer, and standardize the feature component matrix to obtain standardized feature data;

[0092] Based on the standardized characteristic data, a molecular orientation change sequence was constructed. The molecular orientation was increased by 1°, 2°, 3° and decreased by 1°, 2°, 3° on the basis of the baseline value, and 7 sets of molecular orientation change data were obtained.

[0093] Substitute the 7 groups of molecular orientation change data into the birefringence calculation formula n=n0+Δn(3cos2θ-1) / 2, where n is the birefringence, n0 is the isotropic refractive index, Δn is the maximum birefringence, and θ is the molecular orientation angle to obtain each group of birefringence data;

[0094] Jones matrix calculation was performed on each set of birefringence data. The incident light was set to four polarization states of 0°, 45°, 90°, and 135°, and the transmittance values ​​of each set of birefringence data under the four polarization states were calculated to obtain the polarization state-transmittance correspondence data.

[0095] Based on the polarization state-transmittance correspondence data, the transmittance change caused by the change of each group of molecular orientation is calculated, and the central difference method is used to calculate the change rate of transmittance to molecular orientation to obtain the optical performance response curve;

[0096] Integrate the optical performance response curve, calculate the area under the curve in the range of 0°-180° at intervals of 5°, obtain the cumulative impact value of the optical performance of each polarization state, normalize the cumulative impact value of the optical performance according to the maximum value, and perform matrix multiplication operation with the standardized characteristic data to obtain the initial weight data of the polarization state;

[0097] The initial weight data of the polarization state is numerically mapped through the exponential function exp(x) and normalized by the L2 norm to obtain the importance weight of the polarization state of each polarization layer.

[0098] Specifically, singular value decomposition is performed on the fused eigenvector to extract the most important eigencomponents. Singular value decomposition is a matrix decomposition method that decomposes a matrix into the product of three matrices, namely the left singular matrix, the diagonal singular value matrix and the right singular matrix. By performing singular value decomposition on the fused eigenvector, its singular value is obtained, which reflects the amount of information of the eigenvector in different dimensions. By selecting eigencomponents with a cumulative contribution rate greater than 95%, the information that contributes greatly to the prediction of polarizer performance can be effectively retained. Based on the selected eigencomponents, the eigencomponent matrix of each polarizing layer is generated, and these matrices are standardized to obtain standardized eigendata. Based on the standardized eigendata, a molecular orientation change sequence is constructed. Assuming that the initial molecular orientation is the reference value, in order to simulate the change of molecular orientation in the actual process, the reference value is increased or decreased, specifically by increasing by 1°, 2°, 3°, and decreasing by 1°, 2°, 3°, respectively, to obtain 7 groups of molecular orientation change data. For these data, the birefringence corresponding to each group of data is obtained by the birefringence calculation formula. The relationship between birefringence and molecular orientation is nonlinear, so each increase or decrease in orientation will result in different changes in birefringence. Substitute the birefringence data into the Jones matrix calculation. The Jones matrix is ​​a mathematical tool that describes the changes in light as it propagates in a polarizer. Assume that the incident light is set to four polarization states of 0°, 45°, 90°, and 135°, and calculate the transmittance values ​​of each set of birefringence data in the four polarization states. Transmittance is a measure of energy loss after light passes through a polarizer, and it has a close relationship with the polarization state. Through the Jones matrix calculation, the transmittance change caused by the change in birefringence in each polarization state is obtained. The transmittance value is used to construct the corresponding relationship data between the polarization state and the transmittance. Analyze the response of transmittance to changes in molecular orientation. The central difference method is used to calculate the rate of change of transmittance to molecular orientation. The central difference method is a numerical differentiation method, and the rate of change of transmittance is calculated by the following formula:

[0099] ;

[0100] in, The molecular orientation angle is The transmittance at is the increment of molecular orientation. In this way, the transmittance change rate caused by the change of molecular orientation of each group is obtained, and the optical performance response curve is constructed based on this. The optical performance response curve is integrated to obtain the cumulative impact value of the optical performance of each polarizing layer. In the range of 0° to 180, the integration is performed at intervals of 5°, and the area under the curve is calculated. The formula is:

[0101] ;

[0102] in, is the cumulative impact value of optical performance, is the response of transmittance to the rate of change of molecular orientation. The purpose of integration is to sum up the response values ​​at different orientation angles to obtain an overall optical performance index. In this way, the overall impact of each group of molecular orientation changes on the performance of the polarizer is quantified. The cumulative impact values ​​of optical performance are normalized to ensure consistency in the comparison of different groups of data. The standardized characteristic data and the cumulative impact values ​​of optical performance are matrix multiplied to obtain the initial weight data of the polarization state. The effects of different characteristics on the performance of the polarizer are combined to obtain the initial weight of each polarizing layer on the polarization state. Use an exponential function to numerically map the initial weight data of the polarization state. The formula for exponential mapping is:

[0103] ;

[0104] in, is the weight data after mapping, is the initial weight data. This mapping process effectively amplifies the influence of larger weights and reduces the influence of smaller weights, ensuring that the optimization process focuses on the polarization state with greater influence. L2 norm normalization is performed to ensure that the numerical range of the weight data is consistent. The formula is:

[0105] ;

[0106] Through this step, the importance weight of the polarization state of each polarizing layer is finally obtained.

[0107] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0108] Based on the polarization state importance weight and the initial stretch ratio data, a state space matrix is ​​constructed, and the state vector includes three components: the current stretch ratio, the target stretch ratio and the polarization state importance weight, and the initial state data is obtained;

[0109] The initial state data is input into the policy network. The policy network contains three fully connected layers. The first and second layers use the ReLU activation function, and the third layer uses the tanh activation function to limit the action range, and obtain the stretch ratio adjustment action data;

[0110] Discretize the stretch ratio adjustment action data, divide the continuous action space into 21 discrete action points, and the corresponding stretch ratio adjustment range is [-1, 1], the step size is 0.1, and the discretized action data is obtained;

[0111] The discretized action data is input into the value evaluation network. The value evaluation network consists of 4 fully connected layers. Each layer uses the LeakyReLU activation function. The last layer outputs the Q value to obtain the action value data.

[0112] Perform dual advantage function decomposition on the action value data, separate the state value function and the action advantage function, and weight the importance of different actions through the attention mechanism to obtain advantage weighted data;

[0113] Based on advantage-weighted data, a priority experience replay buffer is constructed. The importance score of each action is used as the sampling weight. High-value experience data is selected through a stratified sampling method to obtain optimized trajectory data.

[0114] The optimized trajectory data is input into the target network. The target network structure is the same as the policy network. The network parameter tau=0.01 is updated by soft updating to obtain the target action data.

[0115] The temporal difference error is calculated based on the target motion data and the discretized motion data, and the weighted average is performed in combination with the polarization state importance weight to generate the stretching ratio adjustment strategy for each polarizing layer.

[0116] Specifically, based on the polarization state importance weight and the initial stretch ratio data, a state space matrix is ​​constructed. The state vector contains three components: the current stretch ratio, the target stretch ratio, and the polarization state importance weight. The initial stretch ratio data is obtained by process parameters or experimental measurements, and the polarization state importance weight is obtained through the aforementioned polarization performance analysis steps. These data together constitute the initial input of the state space. Assume that the current stretch ratio is , the target stretch ratio is , the polarization state importance weight is , then the state vector is expressed as:

[0117] ;

[0118] The state vector conveys the importance of the current stretching state of the polarizing layer and its corresponding polarization state. The initial state data is input into the policy network for stretching ratio adjustment. The policy network decides how to adjust the stretching ratio to optimize the performance of the polarizer by learning the mapping relationship between state and action. The policy network consists of three fully connected layers, of which the first two layers use the ReLU activation function and the last layer uses the tanh activation function to limit the range of action. The ReLU activation function has good nonlinear properties, which helps the network capture complex mapping relationships, while the tanh activation function limits the output to the range of [-1,1] to ensure that the output stretching ratio adjustment action is reasonable. The specific network structure is expressed as:

[0119] ;

[0120] in, ,and are the weight matrices of the 1st, 2nd, and 3rd layers respectively, ,and is the bias term, s is the input state vector, The stretch ratio adjustment action data is output. The stretch ratio adjustment action data is discretized. The continuous stretch ratio adjustment range [-1,1] is divided into 21 discrete action points, each action point corresponds to a step size of 0.1, and the adjustment range is discretized into 21 possible stretch ratio adjustment values. Each discrete action data is represented as a discretized action set:

[0121] ;

[0122] Map the continuous decision space into a limited discrete action space to reduce computational complexity and improve the stability of the algorithm. Input the discretized action data into the value evaluation network to evaluate the value of each action. The value evaluation network contains four fully connected layers, each of which uses the LeakyReLU activation function, which can effectively handle negative values ​​in the network and prevent neurons from "dying" during training. The last layer outputs a Q value, which represents the expected return of each action. The specific structure of the value evaluation network is expressed as:

[0123] LeakyReLU LeakyReLU LeakyReLU ;

[0124] in, The Q-value, which represents a given state and action, reflects the expected return of choosing a specific action in the current state. For the calculation result of the Q-value, a double advantage function decomposition is performed to decompose the Q-value into two parts: the state value function and the action advantage function. Through decomposition, the actual value of each action can be evaluated more accurately, avoiding the overestimation problem in the traditional Q-learning method. The state value function represents the average value of all possible actions in a certain state, while the action advantage function represents the advantage of a specific action over other actions in the same state. Through the advantage function, the quality of the action is evaluated and the deviation caused by overestimation of the action value is reduced. On this basis, the attention mechanism is introduced to weight different actions to obtain the advantage weighted data of each action. The attention mechanism allows the network to allocate more computing resources according to the importance of the action, so as to learn and optimize more effectively. The advantage weighted data is expressed by the following formula:

[0125] ;

[0126] in, is the advantage data of the action, is the weighting coefficient based on the attention mechanism. Based on the advantage weighted data, a priority experience replay buffer is constructed. In reinforcement learning, the experience replay buffer is used to store the interaction records of the agent in the environment and update the strategy by sampling these experiences. Through the stratified sampling method, weighted sampling is performed according to the importance score of each experience data, and high-value experience that contributes more to model training is prioritized to obtain optimized trajectory data. The optimized trajectory data is input into the target network. The structure of the target network is the same as that of the policy network, but its parameter update is performed using a soft update method. The soft update method improves the stability of training by smoothly updating the network parameters. The update rule is expressed as:

[0127] ;

[0128] in, are the parameters of the target network, are the parameters of the current network, is the proportional constant of the soft update, usually set to 0.01. Based on the action data output by the target network and the discretized action data, the temporal difference error is calculated and weighted averaged in combination with the polarization state importance weight. The temporal difference error is used to measure the difference between the current strategy and the target strategy, thereby guiding the adjustment of the strategy. The temporal difference error is expressed by the following formula:

[0129] ;

[0130] in, For instant rewards, is the discount factor, is the Q value of the target network, For the next state and action, is the Q value of the current network. Through this step, a stretch ratio adjustment strategy for each polarizing layer is finally generated, which can optimize the performance of the polarizer based on the different importance of the polarization state and the adjustment requirements of the stretch ratio.

[0131] In a specific embodiment, the process of executing step 600 may specifically include the following steps:

[0132] Based on the stretching ratio adjustment strategy, a first process constraint matrix and a second process constraint matrix are constructed, wherein the first process constraint matrix includes a stretching temperature range [120°C, 180°C] and a heating rate range [1°C / min, 5°C / min], and the second process constraint matrix includes a stretching speed range [5mm / min, 50mm / min] and a stretching acceleration range [0.1mm / min2, 1mm / min], and the first constraint condition and the second constraint condition are obtained;

[0133] The first constraint condition and the second constraint condition are collaboratively optimized to construct the target, the first optimization target is set to maximize the polarization efficiency, the second optimization target is set to maximize the uniformity of stress distribution, and the third optimization target is set to minimize the energy consumption, so as to obtain a multi-objective optimization function;

[0134] According to the multi-objective optimization function, a first control variable matrix and a second control variable matrix are constructed, wherein the first control variable matrix describes the temperature-time relationship, and the second control variable matrix describes the speed-displacement relationship, and a global optimization variable space is obtained;

[0135] Based on the global optimization variable space, the segmented optimization solution is performed, and the stretching process is divided into the preheating section, the stretching section and the cooling section. The optimal parameters are calculated for each section to obtain the segmented control parameter sequence.

[0136] The adjacent segments of the segmented control parameter sequence are optimized for transition segments, and a first transition function and a second transition function are introduced, wherein the first transition function controls the continuity of temperature change, and the second transition function controls the continuity of speed change, so as to obtain a smooth transition sequence;

[0137] Based on the smooth transition sequence, a first parameter compensation model and a second parameter compensation model are constructed, wherein the first parameter compensation model is used for temperature lag correction, and the second parameter compensation model is used for speed response correction, to obtain compensation optimization parameters;

[0138] The compensation optimization parameters are converted into control instructions according to the first quantization accuracy and the second quantization accuracy, wherein the first quantization accuracy is the minimum division value of temperature control 0.1°C, and the second quantization accuracy is the minimum division value of speed control 0.1mm / min, to obtain a digital control instruction;

[0139] Based on the digital control instructions, the timing combination is arranged to generate a dual-sequence control scheme including a main control sequence and a backup control sequence. The main control sequence is used for standard working conditions, and the backup control sequence is used for abnormal working condition switching, thereby obtaining the stretching temperature and speed control sequence of each polarizing layer.

[0140] Specifically, according to the requirements of the stretching ratio adjustment strategy, the first process constraint matrix is ​​constructed, which includes the stretching temperature range and the heating rate range, specifically the temperature range [120℃, 180℃] and the heating rate range [1℃ / min, 5℃ / min]. This matrix defines the legal value range of temperature and heating rate to ensure that in the actual production process, the changes in temperature and heating rate will not exceed the safety and performance limits specified by the process. The second process constraint matrix includes the stretching speed range [5mm / min, 50mm / min] and the stretching acceleration range [0.1mm / min², 1mm / min²], which defines the constraints that the stretching speed and acceleration should follow during the stretching process to ensure the stability of the process and the quality of the final product. The specific structure of the first process constraint matrix and the second process constraint matrix is ​​expressed as:

[0141] ;

[0142] Among them, the first and second rows represent the upper and lower limits of temperature and heating rate, respectively, and the second matrix defines the range of stretching speed and acceleration. The first constraint and the second constraint are collaboratively optimized. Three optimization goals are set: maximizing polarization efficiency, by adjusting the control parameters of temperature and speed, the polarization efficiency of the polarizer is optimized; maximizing the uniformity of stress distribution, ensuring the uniform distribution of stress during stretching, and preventing unstable performance of the polarizer caused by uneven stress; minimizing energy consumption, by reasonably setting process parameters, reducing energy consumption in the production process. In order to comprehensively consider these goals, a multi-objective optimization function is constructed as shown below:

[0143] Polarization efficiency Stress inhomogeneity Energy consumption;

[0144] in, are weight coefficients of each objective. By adjusting these coefficients, different optimization objectives are weighted to achieve comprehensive optimization of the process. The first control variable matrix and the second control variable matrix are constructed according to the multi-objective optimization function. The first control variable matrix describes the temperature-time relationship, and can control the temperature change during the stretching process by adjusting the temperature and the heating rate; the second control variable matrix describes the speed-displacement relationship, reflecting the relationship between the stretching speed and the displacement during the stretching process. The control variable matrix is ​​expressed as:

[0145] ;

[0146] in, are the starting and ending values ​​of the temperature, are the start and end values ​​of the time, are the starting and ending values ​​of the speed, , are the starting and ending values ​​of the displacement. The segmented optimization solution is performed based on the global optimization variable space. The entire stretching process is divided into preheating section, stretching section and cooling section, and the optimal control parameters are calculated for each section. By independently optimizing each section, the optimal temperature, speed and displacement control sequence is obtained. For example, in the preheating section, the temperature needs to be heated at a certain heating rate to ensure that the material reaches the ideal stretching temperature; in the stretching section, the speed needs to be controlled within a specific range to ensure that the stretching process of the material proceeds smoothly; in the cooling section, the temperature and speed need to be gradually reduced to avoid stress concentration in the material due to excessive cooling. The control parameter sequence for segmented optimization is expressed as:

[0147] ;

[0148] ;

[0149] ;

[0150] In order to ensure smooth transition between different segments, the first transition function and the second transition function are introduced. The first transition function controls the continuity of temperature change, while the second transition function controls the continuity of speed change. It can be expressed in the following form:

[0151] ;

[0152] ;

[0153] in, and To adjust the parameters of transition speed, and Respectively represent the current temperature and speed, and is the target temperature and speed. Based on the smooth transition sequence, the first parameter compensation model and the second parameter compensation model are constructed to correct the temperature hysteresis and speed response. The temperature hysteresis and speed response deviation will lead to the gap between the actual control and the expected control in the stretching process, affecting the quality of the final product. Through the design of the compensation model, this deviation can be effectively reduced. The compensation model is achieved by introducing the hysteresis correction factor:

[0154] ;

[0155] ;

[0156] The compensation optimization parameters are quantified, and the control instructions are converted according to the quantization accuracy. For example, the minimum division value of temperature control is 0.1°C, and the minimum division value of speed control is 0.1mm / min. The parameters after the control instruction conversion are used in the digital control system to ensure the precise control of process parameters in the actual production process. Through the timing combination arrangement, a dual-sequence control scheme including the main control sequence and the backup control sequence is generated. The main control sequence is suitable for standard working conditions, while the backup control sequence is used to cope with switching under abnormal working conditions. Through the above steps, it is ensured that the polarizer can stably achieve the best polarization effect in various production environments.

[0157] The above describes the performance optimization design method of the multilayer composite polarizer in the embodiment of the present application. The following describes the performance optimization design system 10 of the multilayer composite polarizer in the embodiment of the present application. Figure 2 In one embodiment of the present application, a system 10 for optimizing the performance of a multilayer composite polarizer includes:

[0158] The analysis module 11 is used to perform polarization spectrum analysis on the use environment of the display screen, obtain incident polarization spectrum data, and calculate the initial stretch ratio data of each polarizing layer;

[0159] The measuring module 12 is used to measure the production process data of each polarizing layer of the sample product respectively, and obtain the molecular orientation data and polarization performance data of each polarizing layer;

[0160] A fusion module 13 is used to input the molecular orientation data and polarization performance data of each polarizing layer into the dual-domain and global context feature extraction networks for feature extraction and fusion, and obtain a fused feature vector of each polarizing layer;

[0161] A calculation module 14 is used to calculate the influence of the change of the molecular orientation degree of each polarizing layer on the optical properties of different polarization states based on the fused feature vector, and obtain the importance weight of the polarization state of each polarizing layer;

[0162] An optimization module 15 is used to perform reinforcement learning optimization on the initial stretch ratio data according to the polarization state importance weight, and generate a stretch ratio adjustment strategy for each polarizing layer;

[0163] The generation module 16 is used to perform multi-objective optimization and sequence quadratic programming on the processing parameters of each polarizing layer based on the stretching ratio adjustment strategy, and generate a stretching temperature and speed control sequence for each polarizing layer.

[0164] Through the synergy of the above-mentioned components, by introducing dual-domain and global context feature extraction networks for feature extraction and fusion, combined with reinforcement learning optimization and multi-objective sequential quadratic programming, precise optimization of polarizer performance and adaptive control of process parameters are achieved. This method has three outstanding technical advantages: through dual-path feature extraction in the spatial domain and frequency domain, combined with global context feature encoding, it realizes the comprehensive capture of molecular orientation and polarization performance characteristics; adopts a reinforcement learning method based on the importance weight of polarization state, establishes an adaptive optimization mechanism for stretching ratio parameters, and makes the adjustment of process parameters more accurate; through multi-objective optimization and sequential quadratic programming, combined with segmented control and transition optimization strategies, it realizes precise control of stretching temperature and speed, and enhances the robustness of the process through the main and standby control sequence switching mechanism. The present invention significantly improves the performance indicators and production efficiency of polarizers.

[0165] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a system bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0166] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned performance optimization design methods for the multi-layer composite polarizer.

[0167] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0168] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned performance optimization design methods for the multi-layer composite polarizer.

[0169] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the present application scheme, and does not constitute a limitation on the electronic device 300 involved in the present application scheme. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0170] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0171] It should be noted that those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the performance optimization design method of the aforementioned multi-layer composite polarizer, and will not be repeated here.

[0172] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the performance optimization design method of the multilayer composite polarizer provided in the embodiment of the present application.

[0173] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0174] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0175] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0176] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A performance optimization design method for a multi-layer composite polarizer, characterized in that: The method comprises: Perform polarization spectrum analysis on the use environment of the display screen, obtain incident polarization spectrum data, and calculate the initial stretch ratio data of each polarizing layer; The production process data of each polarizing layer of the sample product is measured respectively to obtain the molecular orientation data and polarization performance data of each polarizing layer; The molecular orientation data and polarization performance data of each polarizing layer are respectively input into the dual-domain and global context feature extraction networks for feature extraction and fusion, and the fused feature vector of each polarizing layer is obtained; Based on the fused feature vector, the influence of the change of the molecular orientation degree of each polarizing layer on the optical properties of different polarization states is calculated to obtain the importance weight of the polarization state of each polarizing layer; According to the polarization state importance weight, the initial stretch ratio data is optimized by reinforcement learning to generate a stretch ratio adjustment strategy for each polarizing layer; Based on the stretching ratio adjustment strategy, multi-objective optimization and sequence quadratic programming are performed on the processing parameters of each polarizing layer to generate a stretching temperature and speed control sequence for each polarizing layer.

2. The performance optimization design method of the multi-layer composite polarizer according to claim 1, characterized in that: The polarization spectrum analysis of the use environment of the display screen is performed to obtain incident polarization spectrum data, and the initial stretch ratio data of each polarizing layer is calculated, including: Perform multi-angle light intensity measurement on the natural light source in the use environment of the display screen to obtain spatial distribution data of incident light intensity, and perform polarization state decomposition on the spatial distribution data of incident light intensity to obtain horizontal polarization component data and vertical polarization component data; Performing spectral analysis on the horizontal polarization component data and the vertical polarization component data to obtain wavelength distribution data of each polarization state, and constructing a polarization light transmittance optimization objective function based on the wavelength distribution data to obtain a polarization light optimization weight matrix; Constructing an optical performance transfer equation of a multilayer composite polarizer according to the polarized light optimization weight matrix to obtain a polarization efficiency target value of each polarizing layer, and performing stress-strain analysis on the polarization efficiency target value to obtain molecular orientation data required for each polarizing layer; The material mechanical property analysis is performed based on the molecular orientation degree data to obtain stress distribution data of each polarizing layer, and the stretching process parameter calculation is performed based on the stress distribution data to obtain initial stretching ratio data of each polarizing layer.

3. The performance optimization design method of the multi-layer composite polarizer according to claim 2, characterized in that: The production process data of each polarizing layer of the sample product is measured respectively to obtain the molecular orientation data and polarization performance data of each polarizing layer, including: The multi-point polarization spectrum analyzer is set to a wavelength scanning range of 450-750nm, and the transmittance of the sample product is measured to obtain the light intensity distribution data of each polarizing layer at different wavelengths; Perform polarization state separation calculation based on the light intensity distribution data to obtain parallel transmittance data and vertical transmittance data; Inputting the parallel transmittance data and the vertical transmittance data into an extinction ratio calculation model to obtain polarization performance data of each polarizing layer; Performing an ellipsometer incident angle scan on the sample product to obtain amplitude ratio data and phase difference data of each polarizing layer, and performing birefringence calculation based on the amplitude ratio data and the phase difference data to obtain refractive index anisotropy data of each polarizing layer; Performing molecular orientation inversion calculation on the refractive index anisotropy data, and simulating polarization performance of each polarizing layer based on the molecular orientation data to obtain theoretical polarization performance data; The theoretical polarization performance data is compared and analyzed with the polarization performance data to determine the molecular orientation degree data of each polarizing layer.

4. The performance optimization design method of the multi-layer composite polarizer according to claim 3, characterized in that: The molecular orientation data and polarization performance data of each polarizing layer are respectively input into the dual-domain and global context feature extraction networks for feature extraction and fusion to obtain a fused feature vector of each polarizing layer, including: Inputting the molecular orientation data into a spatial domain feature extraction branch, the spatial domain feature extraction branch comprises four convolutional layers, each of which uses a convolution operation with a step size of 2 and BatchNorm normalization, and uses a LeakyReLU activation function to obtain a spatial domain feature map; The molecular orientation data is subjected to fast Fourier transform and input into a frequency domain feature extraction branch, wherein the frequency domain feature extraction branch comprises four convolutional layers, each of which uses a dilated convolution and SE attention mechanism to obtain a frequency domain feature map; Performing channel attention weighting on the spatial domain feature map and the frequency domain feature map, and performing feature fusion through a cross-domain feature fusion module to obtain a dual-domain feature vector; Inputting the polarization performance data into a global context encoder, wherein the global context encoder comprises three multi-head self-attention layers, each of which comprises eight attention heads, to obtain a global context feature vector; Performing multi-scale feature decomposition on the global context feature vector, extracting context information of different scales through three parallel 1×1 convolutional layers to obtain multi-scale context features; Inputting the dual-domain feature vector and the multi-scale context feature into a feature fusion network, wherein the feature fusion network comprises two fully connected layers and one residual connection, to obtain an initial fusion feature; A self-calibration operation is performed on the initial fused features, and long-range dependencies are captured through a non-local attention module to obtain calibrated fused features. The calibrated fused features are input into a feature compression network, and a fused feature vector of each polarization layer is obtained through two fully connected layers and a Dropout layer.

5. The performance optimization design method of the multi-layer composite polarizer according to claim 4, characterized in that: The method of calculating the influence of the change of the molecular orientation degree of each polarizing layer on the optical properties of different polarization states based on the fused feature vector to obtain the importance weight of the polarization state of each polarizing layer includes: Performing a singular value decomposition operation on the fused feature vector, selecting feature components with a cumulative contribution rate greater than 95%, generating a feature component matrix for each polarizing layer, and performing a standardization process on the feature component matrix to obtain standardized feature data; Based on the standardized characteristic data, a molecular orientation change sequence is constructed, and the molecular orientation is increased by 1°, 2°, 3° and decreased by 1°, 2°, 3° on the basis of the reference value, to obtain 7 sets of molecular orientation change data; Substitute the 7 groups of molecular orientation change data into the birefringence calculation formula n=n0+Δn(3cos2θ-1) / 2, where n is the birefringence, n0 is the isotropic refractive index, Δn is the maximum birefringence, and θ is the molecular orientation angle to obtain each group of birefringence data; Jones matrix calculation is performed on each group of birefringence data, and the incident light is set to four polarization states of 0°, 45°, 90°, and 135°, respectively, and the transmittance values ​​of each group of birefringence data under the four polarization states are calculated to obtain polarization state-transmittance correspondence relationship data; Based on the polarization state-transmittance correspondence data, the transmittance change caused by the change in molecular orientation of each group is calculated, and the change rate of transmittance to molecular orientation is calculated by using the central difference method to obtain an optical performance response curve; Performing an integration operation on the optical performance response curve, calculating the area under the curve at intervals of 5° within the range of 0°-180°, obtaining the cumulative impact value of the optical performance of each polarization state, normalizing the cumulative impact value of the optical performance according to the maximum value, and performing a matrix multiplication operation with the standardized characteristic data to obtain the initial weight data of the polarization state; The polarization state initial weight data is numerically mapped through an exponential function exp(x), and L2 norm normalization is performed to obtain the polarization state importance weight of each polarizing layer.

6. The performance optimization design method of the multi-layer composite polarizer according to claim 5, characterized in that: The step of performing reinforcement learning optimization on the initial stretch ratio data according to the polarization state importance weight to generate a stretch ratio adjustment strategy for each polarizing layer includes: Based on the polarization state importance weight and the initial stretch ratio data, a state space matrix is ​​constructed, wherein the state vector includes three components: the current stretch ratio, the target stretch ratio and the polarization state importance weight, and the initial state data is obtained; Inputting the initial state data into a strategy network, wherein the strategy network comprises three fully connected layers, wherein the first and second layers use a ReLU activation function, and the third layer uses a tanh activation function to limit the action range, thereby obtaining stretch ratio adjustment action data; Discretization processing is performed on the stretching ratio adjustment action data, and the continuous action space is divided into 21 discrete action points, and the corresponding stretching ratio adjustment range is [-1, 1], and the step length is 0.1, so as to obtain discretized action data; The discretized action data is input into a value evaluation network, wherein the value evaluation network comprises four fully connected layers, each layer uses a LeakyReLU activation function, and the last layer outputs a Q value to obtain action value data; Performing dual advantage function decomposition on the action value data, separating the state value function and the action advantage function, and weighting the importance of different actions through an attention mechanism to obtain advantage weighted data; Based on the advantage weighted data, a priority experience replay buffer is constructed, the importance score of each action is used as a sampling weight, and high-value experience data is selected through a stratified sampling method to obtain optimized trajectory data; The optimized trajectory data is input into the target network, the target network structure is the same as the policy network, and the network parameter tau=0.01 is updated by soft updating to obtain the target action data; A time difference error is calculated based on the target motion data and the discretized motion data, and a weighted average is performed in combination with the polarization state importance weight to generate a stretch ratio adjustment strategy for each polarizing layer.

7. The performance optimization design method of the multi-layer composite polarizer according to claim 6, characterized in that: The multi-objective optimization and sequence quadratic programming of the processing parameters of each polarizing layer based on the stretching ratio adjustment strategy are performed to generate a stretching temperature and speed control sequence for each polarizing layer, including: Based on the stretching ratio adjustment strategy, a first process constraint matrix and a second process constraint matrix are constructed, wherein the first process constraint matrix includes a stretching temperature range [120°C, 180°C] and a heating rate range [1°C / min, 5°C / min], and the second process constraint matrix includes a stretching speed range [5mm / min, 50mm / min] and a stretching acceleration range [0.1mm / min2, 1mm / min], to obtain a first constraint condition and a second constraint condition; Constructing collaborative optimization objectives for the first constraint condition and the second constraint condition, setting the first optimization objective to maximize polarization efficiency, the second optimization objective to maximize stress distribution uniformity, and the third optimization objective to minimize energy consumption, to obtain a multi-objective optimization function; Constructing a first control variable matrix and a second control variable matrix according to the multi-objective optimization function, wherein the first control variable matrix describes the temperature-time relationship, and the second control variable matrix describes the speed-displacement relationship, to obtain a global optimization variable space; Based on the global optimization variable space, a segmented optimization solution is performed, the stretching process is divided into a preheating section, a stretching section and a cooling section, and the optimal parameters are calculated for each section to obtain a segmented control parameter sequence; Performing transition segment optimization on adjacent segments of the segmented control parameter sequence, introducing a first transition function and a second transition function, wherein the first transition function controls the continuity of temperature change, and the second transition function controls the continuity of speed change, to obtain a smooth transition sequence; Based on the smooth transition sequence, a first parameter compensation model and a second parameter compensation model are constructed, wherein the first parameter compensation model is used for temperature lag correction, and the second parameter compensation model is used for speed response correction, to obtain compensation optimization parameters; The compensation optimization parameters are converted into control instructions according to a first quantization accuracy and a second quantization accuracy, wherein the first quantization accuracy is a minimum division value of 0.1°C for temperature control, and the second quantization accuracy is a minimum division value of 0.1mm / min for speed control, to obtain a digital control instruction; Based on the digital control instructions, a timing combination is arranged to generate a dual-sequence control scheme including a main control sequence and a backup control sequence, wherein the main control sequence is used for standard working conditions and the backup control sequence is used for abnormal working condition switching, thereby obtaining the stretching temperature and speed control sequence of each polarizing layer.

8. A performance optimization design system for a multi-layer composite polarizer, characterized in that: A method for optimizing the performance of a multilayer composite polarizer according to any one of claims 1 to 7, the system comprising: An analysis module is used to perform polarization spectrum analysis on the use environment of the display screen, obtain incident polarization spectrum data, and calculate the initial stretch ratio data of each polarizing layer; The measurement module is used to measure the production process data of each polarizing layer of the sample product to obtain the molecular orientation data and polarization performance data of each polarizing layer; A fusion module, used to input the molecular orientation data and polarization performance data of each polarizing layer into the dual-domain and global context feature extraction networks for feature extraction and fusion, and obtain a fused feature vector of each polarizing layer; A calculation module, for calculating the influence of the change of the molecular orientation degree of each polarizing layer on the optical properties of different polarization states based on the fused feature vector, and obtaining the importance weight of the polarization state of each polarizing layer; An optimization module, configured to perform reinforcement learning optimization on the initial stretch ratio data according to the polarization state importance weight, and generate a stretch ratio adjustment strategy for each polarizing layer; A generation module is used to perform multi-objective optimization and sequence quadratic programming on the processing parameters of each polarizing layer based on the stretching ratio adjustment strategy, and generate a stretching temperature and speed control sequence for each polarizing layer.

9. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the performance optimization design method of the multi-layer composite polarizer according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the performance optimization design method of the multilayer composite polarizer according to any one of claims 1 to 7 is implemented.

Images