Cooperative control method for surface stress distribution and optical performance of high-color-gamut peep-proof film

Abstract

The invention provides a high-color-gamut peep-proof film surface stress distribution and optical performance cooperative control method, which relates to the technical field of optical engineering, and comprises the following steps: acquiring peep-proof film multilayer structure data, establishing a feature set, forming a cooperative representation vector by using deep neural network joint coding, and generating a cooperative control strategy based on a bidirectional influence weight matrix. And making a process parameter adjustment scheme. According to the invention, collaborative optimization of stress regulation and optical performance is realized, the color rendition degree and visual angle control precision of the peep-proof film are improved, and the service life of the product is prolonged.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical engineering, in particular to a high color gamut privacy film surface stress distribution and optical performance collaborative control method. BACKGROUND

[0002] The high color gamut privacy film is a functional optical film applied to display devices, which can effectively prevent information leakage outside the viewing angle while ensuring display color reproduction. The high color gamut privacy film is usually composed of multiple functional film layers, including a base film layer, a privacy microstructure layer, an anti-reflection layer, a hardening layer, etc. Its working principle is to control the light transmission path through special microstructure design to realize light transmission within a specific viewing angle range and block light at other viewing angles.

[0003] In the manufacturing process of traditional privacy films, stress distribution and optical performance are often treated as two independent technical problems, which have the following shortcomings: there is a lack of effective stress and optical performance correlation analysis method, most studies treat stress control and optical design as independent tasks, ignoring the complex interaction between the two, which makes it difficult to achieve optimal comprehensive performance of the final product. The traditional stress control method mainly relies on empirical formulas and simplified models, which cannot accurately predict the stress distribution state in the multi-layer composite structure and its evolution law over time and environment, resulting in performance instability or even failure of the privacy film in actual use. The existing optical compensation technology fails to fully consider the changes in optical performance caused by stress, and the compensation scheme is often static, lacking the ability to adaptively adjust the dynamic evolution of stress, making it difficult to achieve optical performance stability of the high color gamut privacy film throughout its life cycle. SUMMARY

[0004] The embodiment of the present application provides a high color gamut privacy film surface stress distribution and optical performance collaborative control method, which can solve the problems in the prior art.

[0005] In a first aspect, the embodiment of the present application provides a high color gamut privacy film surface stress distribution and optical performance collaborative control method, which comprises: obtaining multi-layer structure data of the privacy film, constructing a stress state feature set and an optical performance feature set; jointly encoding the stress state feature set and the optical performance feature set through a deep neural network to obtain a stress-optical collaborative representation vector, the collaborative representation vector simultaneously embedding stress evolution trend information and optical response sensitivity information in a unified feature space, identifying the interactive dependency relationship between the stress state features and the optical performance features through an attention mechanism, establishing a bidirectional influence weight matrix, the bidirectional influence weight matrix representing the forward influence path weight of stress evolution on optical performance and the reverse constraint path weight of optical optimization demand on stress distribution; input the synergistic characterization vector into a strategy generation network, the strategy generation network analyzes a forward influence path of stress regulation on optical performance and a reverse constraint path of optical optimization on stress distribution based on the bidirectional influence weight matrix, and generates a synergistic control strategy including stress regulation instructions and optical compensation instructions; generate a process parameter adjustment scheme according to the synergistic control strategy.

[0006] The steps of obtaining the multi-layer structure data of the privacy protection film, and constructing the stress state feature set and the optical performance feature set include: The material type and thickness information of each functional layer are respectively converted into physical property parameters and optical parameters, the interlayer stress gradient and stress distribution uniformity are calculated based on the physical property parameters to constitute the stress state feature set, and the light intensity distribution and color gamut coverage range under different viewing angles are calculated based on the optical parameters to extract the privacy protection viewing angle response feature and the color gamut boundary feature to constitute the optical performance feature set.

[0007] The steps of jointly encoding the stress state feature set and the optical performance feature set by a deep neural network to obtain a stress-optical synergistic characterization vector include: A synergistic state mapping space with stress distribution state and optical performance state as two dimensions is constructed, and a state trajectory point of historical iterations is recorded; a synergistic optimal region is determined based on a point set in which stress deviation and optical performance deviation simultaneously satisfy threshold conditions; The stress evolution trend information is extracted by time series unfolding of the stress state feature set, and the optical response sensitivity information is extracted by perturbation test of the optical performance feature set; The stress state feature set and the optical performance feature set are jointly encoded by an encoding layer of a deep neural network, and the stress evolution trend information and the optical response sensitivity information are embedded to obtain an initial synergistic characterization vector; The initial synergistic characterization vector is mapped to the synergistic state mapping space to obtain a current state point, and a deviation degree of the current state point from the synergistic optimal region is calculated; when the deviation degree exceeds a synergistic tolerance threshold, the initial synergistic characterization vector is decomposed into a stress independent component and an optical independent component, reverse compensation is applied according to the deviation contribution degree of each independent component, and then the initial synergistic characterization vector is fused to obtain a synergistic characterization vector.

[0008] The steps of calculating the deviation degree of the current state point from the synergistic optimal region, and when the deviation degree exceeds a synergistic tolerance threshold, decomposing the initial synergistic characterization vector into a stress independent component and an optical independent component, applying reverse compensation according to the deviation contribution degree of each independent component, and then fusing the initial synergistic characterization vector to obtain a synergistic characterization vector include: calculating a Euclidean distance from the current state point to an ideal cooperative state reference point of the cooperative optimal region as a preliminary deviation degree; when the current state point is located outside the boundary of the cooperative optimal region, calculating a shortest penetration distance to the boundary, and taking a weighted sum of the preliminary deviation degree and the penetration distance as a final deviation degree; when the final deviation degree exceeds a cooperative tolerance threshold, decomposing the initial cooperative characterization vector into statistically independent stress independent components and optical independent components by an independent component analysis algorithm; decomposing the final deviation degree into a stress dimension deviation component and an optical dimension deviation component, and calculating a contribution proportion of each independent component to the corresponding dimension deviation component; applying a compensation vector pointing to a stress subspace projection of the ideal cooperative state reference point to the stress independent components, and applying a compensation vector pointing to an optical subspace projection of the ideal cooperative state reference point to the optical independent components, a compensation amplitude being modulated by the contribution proportion of each independent component; and fusing the compensated stress independent components and optical independent components to obtain a cooperative characterization vector.

[0009] The steps of establishing a bidirectional influence weight matrix by recognizing the interactive dependence relationship between the stress state features and the optical performance features through the attention mechanism include: converting the stress state feature set into a query vector, converting the optical performance feature set into a key vector and a value vector; calculating an association strength through inner product operation of the query vector and the key vector, weighting and aggregating the value vector by taking the association strength as the weight to obtain a feature representation containing the interactive dependence relationship, and constructing an interactive dependence matrix; identifying a positive influence path of the stress state features on the optical performance features from the interactive dependence matrix, the positive influence path representing a transmission relationship of stress evolution to optical performance change, and calculating a transmission weight coefficient of each feature node in the positive influence path; identifying a reverse constraint path of the optical performance features on the stress state features from the interactive dependence matrix, the reverse constraint path representing a feedback relationship of optical optimization demand to stress distribution adjustment, and calculating a feedback weight coefficient of each feature node in the reverse constraint path; constructing a bidirectional influence weight matrix based on the transmission weight coefficient and the feedback weight coefficient.

[0010] The steps of inputting the cooperative characterization vector into a strategy generation network, the strategy generation network analyzing a positive influence path of stress regulation on optical performance and a reverse constraint path of optical optimization on stress distribution based on the bidirectional influence weight matrix, and generating a cooperative control strategy including stress regulation instructions and optical compensation instructions include: The collaborative representation vector is input into the policy generation network. Based on the stress evolution trend information and optical response sensitivity information embedded in the collaborative representation vector, the stress distribution deviation vector and optical performance deviation vector between the current state and the target state are analyzed. Based on the transfer weight coefficients in the bidirectional influence weight matrix, the influence contribution of the stress distribution deviation vector on each optical performance feature is traced to construct a path diagram of the effect of stress regulation on optical performance; based on the feedback weight coefficients in the bidirectional influence weight matrix, the constraint strength of the optical performance deviation vector on each stress state feature is traced to construct a path diagram of the reverse constraint of optical optimization on stress distribution. Stress control nodes whose influence contribution exceeds the contribution threshold are extracted from the action path diagram, and stress control instructions are generated according to the weight distribution of the stress control nodes; optical compensation nodes whose constraint strength exceeds the constraint threshold are extracted from the reverse constraint path diagram, and optical compensation instructions are generated according to the weight distribution of the optical compensation nodes. Conflict detection is performed on stress control commands and optical compensation commands. When antagonistic influences are detected, priority ranking and coordination correction are performed based on the path weights of the action path diagram and the reverse constraint path diagram to obtain a cooperative control strategy.

[0011] The steps for conflict detection between stress control commands and optical compensation commands, and prioritization and coordination correction based on path weights in the action path diagram and reverse constraint path diagram when antagonistic influences are detected, include: Construct an influence matrix for stress control commands and optical compensation commands, where each element represents the mutual influence coefficient between corresponding command pairs; identify command pairs with negative mutual influence coefficients from the influence matrix and mark them as conflicting command pairs; For each conflicting command pair, extract the path containing the stress control command in the conflicting command pair from the action path diagram, and calculate the cumulative path weight of the path; extract the path containing the optical compensation command in the conflicting command pair from the reverse constraint path diagram, and calculate the cumulative path weight of the path; compare the cumulative path weights of the two paths, and set the command with the larger cumulative path weight as the high priority command. The lower priority instruction in the conflicting instruction pair is attenuated by a factor that is determined by the ratio of the cumulative path weight of the higher priority instruction to the cumulative path weight of the lower priority instruction.

[0012] This invention establishes a bidirectional mapping relationship between stress state and optical performance, achieving coordinated control of stress distribution and optical performance during the design of high color gamut privacy films, thus avoiding performance conflicts caused by single optimization in traditional methods. By employing deep neural networks and attention mechanisms to jointly encode the stress state feature set and the optical performance feature set, a coordinated representation vector in a unified feature space is obtained. This vector accurately identifies the influence path of stress evolution on optical performance and the constraints of optical optimization on stress distribution, improving the optical performance and mechanical stability of high color gamut privacy films. Furthermore, a strategy generation network generates a coordinated control strategy based on a bidirectional influence weight matrix, including stress adjustment instructions and optical compensation instructions. This makes process parameter adjustments more precise and efficient, significantly improving the overall performance of high color gamut privacy films and extending product lifespan. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the method for synergistic control of surface stress distribution and optical performance of a high color gamut privacy film according to an embodiment of the present invention. Figure 2 A flowchart for generating and handling conflicts in the collaborative control strategy of high color gamut privacy films. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0016] Figure 1 This is a flowchart illustrating the method for synergistic control of surface stress distribution and optical performance of a high color gamut privacy film according to an embodiment of the present invention. Figure 1 As shown, the method includes: Obtain multilayer structural data of privacy films, and construct stress state feature sets and optical performance feature sets; The stress state feature set and the optical performance feature set are jointly encoded by a deep neural network to obtain a stress-optics co-representation vector. This co-representation vector embeds stress evolution trend information and optical response sensitivity information in a unified feature space. An attention mechanism is used to identify the interaction dependency between stress state features and optical performance features, and a bidirectional influence weight matrix is ​​established. This bidirectional influence weight matrix represents the positive influence path weight of stress evolution on optical performance and the negative constraint path weight of optical optimization requirements on stress distribution. The collaborative representation vector is input into the policy generation network, which analyzes the positive influence path of stress modulation on optical performance and the negative constraint path of optical optimization on stress distribution based on the bidirectional influence weight matrix, and generates a collaborative control policy including stress modulation instructions and optical compensation instructions. A process parameter adjustment scheme is generated based on the aforementioned collaborative control strategy.

[0017] In one optional implementation, the steps of acquiring multilayer structure data of the privacy film and constructing a stress state feature set and an optical performance feature set include: The material type and thickness information of each functional layer are converted into physical property parameters and optical parameters, respectively. Based on the physical property parameters, the interlayer stress gradient and stress distribution uniformity are calculated to form a stress state feature set. Based on the optical parameters, the light intensity distribution and color gamut coverage under different viewing angles are calculated, and the privacy viewing angle response features and color gamut boundary features are extracted to form an optical performance feature set.

[0018] For example, it is necessary to collect basic data on the multi-layer structure of the privacy film, including the material type and thickness information of each functional layer. Taking a five-layer privacy film as an example, its composition includes: a surface protective layer (PET material, 25 micrometers thick), an upper adhesive layer (acrylate adhesive, 15 micrometers thick), a microprism array layer (PMMA material, 100 micrometers thick), a lower adhesive layer (acrylate adhesive, 15 micrometers thick), and a bottom substrate (PET material, 50 micrometers thick).

[0019] For each collected material type, the system queries a pre-set material database and converts it into corresponding physical property parameters. These parameters include elastic modulus, Poisson's ratio, coefficient of thermal expansion, and density. For example, PET material has an elastic modulus of approximately 2.5 GPa, a Poisson's ratio of 0.37, and a coefficient of thermal expansion of 7 × 10⁻⁶. -5 / K, density is 1.38 g / cm³ 3 The elastic modulus of PMMA material is approximately 3.0 GPa, Poisson's ratio is 0.35, and the coefficient of thermal expansion is 9 × 10⁻⁶. -5 / K, density is 1.18 g / cm³ 3The elastic modulus of the acrylic resin is approximately 0.5 MPa, Poisson's ratio is 0.45, and the coefficient of thermal expansion is 2 × 10⁻⁶. -4 / K, density is 1.05 g / cm³ 3 .

[0020] Simultaneously, the various material types are converted into optical parameters. These optical parameters mainly include refractive index, transmittance, absorption coefficient, and scattering characteristics. For example, PET material has a refractive index of approximately 1.58 and a transmittance exceeding 90% in the visible light range; PMMA material has a refractive index of approximately 1.49 and a transmittance of approximately 92%; and acrylic resin has a refractive index of approximately 1.47 and a transmittance of approximately 95%. In addition to the basic optical parameters, the geometric parameters of the microprism array layer's microstructure also need to be recorded, including a prism angle of 60°, a prism spacing of 50 micrometers, and a linear array arrangement.

[0021] Based on the acquired physical property parameters, the calculation of interlayer stress gradient needs to consider the differences in elastic modulus and thermal expansion coefficients of adjacent functional layer materials, as well as the actual operating environment temperature changes. Taking the above five-layer structure as an example, under the condition of temperature rising from 20°C to 60°C, the stress gradient between the upper adhesive layer and the microprism array layer is 0.58 MPa / μm, and the stress gradient between the microprism array layer and the lower adhesive layer is 0.62 MPa / μm. The uniformity of stress distribution is characterized by calculating the standard deviation of stress values ​​throughout the multilayer structure. In this example, the standard deviation is 0.15 MPa, indicating that the stress distribution is relatively uniform. The stress state feature set also includes data such as the location of maximum stress and stress concentration factor, recorded in the form of a data vector for easy processing by subsequent machine learning algorithms.

[0022] Based on the acquired optical parameters, a ray tracing algorithm was used to calculate the light intensity distribution at different viewing angles. The incident light was set to be perpendicular to the front, and the observation angle varied from 0° to 90°, with a data point recorded every 5°. During ray tracing, the refraction and reflection of light at the interlayer interfaces were calculated based on the refractive index parameters of each layer, and the attenuation of light in each layer was calculated based on the absorption coefficient parameters. Particularly in the microprism array layer, the deflection path and exit angle distribution of light were calculated using geometric optics laws based on geometric parameters such as a prism angle of 60° and a prism spacing of 50 micrometers. In this example, the transmittance of the privacy film remained above 85% within a viewing angle range of 0° to 30°, but decreased rapidly beyond 35°, dropping to 45% at 40°, 25% at 45°, 10% at 50°, and below 5% at angles of 60° and above. The transmittance data corresponding to different viewing angles are used to construct a viewpoint-transmittance curve feature vector, and the viewing angle corresponding to the transmittance dropping to 50% is marked as the privacy protection critical angle. In this example, the privacy protection critical angle is 35°, which is used as the privacy protection viewing angle response feature.

[0023] The color gamut coverage was calculated using the spectral transfer matrix method. A wavelength-dependent transfer matrix was constructed based on the transmittance and absorption coefficient parameters of each layer. The spectral intensity distribution of the standard RGB three channels (red peak wavelength 630nm, green peak wavelength 530nm, and blue peak wavelength 470nm) after passing through the privacy film at different viewing angles was calculated. Then, the transmitted spectral intensity was converted to the CIE 1976 Lab* color space, and the boundary of the color range that could be presented in this space was calculated. At a normal viewing angle (0°-30°), the calculated color gamut boundary covered 96% of the sRGB standard color gamut; however, at a 45° viewing angle, due to the stronger selective deflection of short-wavelength light by the microprism array, the transmittance of the blue channel (470nm) and green channel (530nm) decreased to 40% and 55% of their original values, respectively, causing the color gamut boundary to shrink to 35% of the sRGB standard. Extract the coordinate point set of the color gamut boundary in Lab* space under each viewing angle, calculate the ratio of the current color gamut area to the sRGB standard color gamut area as the color gamut coverage, and calculate the offset of the color gamut center point as the color distortion. Summarize parameters such as the viewing angle-transmittance curve feature vector, privacy protection critical angle, viewing angle-color gamut coverage curve, and color distortion vector to form an optical performance feature set.

[0024] This invention achieves a precise description of the multilayer structure characteristics of privacy films by converting material type and thickness information into physical and optical parameters, and can accurately construct stress state feature sets and optical performance feature sets.

[0025] In one optional implementation, the step of jointly encoding the stress state feature set and the optical performance feature set using a deep neural network to obtain a stress-optics co-representation vector includes: A collaborative state mapping space with stress distribution state and optical performance state as two dimensions is constructed, and the state trajectory points of historical iterations are recorded; the collaborative optimal region is determined based on the set of points in the state trajectory points where stress deviation and optical performance deviation simultaneously satisfy the threshold condition. The stress state feature set is subjected to time-series unfolding to extract stress evolution trend information; the optical performance feature set is subjected to perturbation testing to extract optical response sensitivity information. The stress state feature set and the optical performance feature set are jointly encoded by the encoding layer of a deep neural network, and the stress evolution trend information and optical response sensitivity information are embedded to obtain an initial collaborative characterization vector. The initial collaborative representation vector is mapped to the collaborative state mapping space to obtain the current state point, and the deviation between the current state point and the collaborative optimal region is calculated. When the deviation exceeds the collaborative tolerance threshold, the initial collaborative representation vector is decomposed into stress-independent components and optical-independent components. After applying reverse compensation based on the deviation contribution of each independent component, the components are re-fused to obtain the collaborative representation vector.

[0026] For example, when constructing the collaborative state mapping space, a two-dimensional coordinate system is established with stress distribution state and optical performance state as two orthogonal dimensions. The stress distribution state dimension uses the standard deviation of stress distribution uniformity as a metric, with a value ranging from 0 to 0.5 MPa. The optical performance state dimension uses a weighted comprehensive score of color gamut coverage and privacy protection critical angle, with color gamut coverage weighted at 0.6 and privacy protection critical angle weighted at 0.4. The comprehensive score is normalized to the range of 0 to 100. After each control strategy is executed, the current stress distribution state and optical performance state are mapped to state points in this two-dimensional space, and the coordinate values ​​are recorded. All state points generated in historical iterations constitute a state trajectory, which is stored in the state history database. The determination of the collaborative optimal region depends on the screening of historical trajectory points. The system sets the stress deviation threshold to 0.12 MPa and the optical performance deviation threshold to 5 points. The target stress distribution state is set with a standard deviation of 0.10 MPa, and the target optical performance comprehensive score is set to 92 points. All trajectory points in the database are traversed, and trajectory points with stress deviations not exceeding 0.12 MPa and optical performance deviations not exceeding 5 points are selected as candidate point sets. Convex hull calculation is performed on the candidate point set to obtain the smallest convex polygon region containing all candidate points as the cooperative optimal region.

[0027] Stress evolution trend information is extracted by temporal unfolding of the stress state feature set. The stress state feature set of the current moment and the previous five consecutive sampling moments are arranged in chronological order, with each sampling period being 10 seconds. A sliding time window technique is used, with a window length of 5 sampling periods. The linear fitting slope of the stress gradient data within the window is calculated; a positive slope indicates an upward trend, and a negative slope indicates a downward trend, with the absolute value reflecting the rate of change. The variance of the stress distribution uniformity data is calculated; a larger variance indicates poor stability. The fitting slope, variance, and the distance of the maximum stress location movement are used as the three core features of stress evolution trend information. For example, the stress gradients of the first five sampling periods are 0.58, 0.60, 0.63, 0.65, and 0.67 MPa / μm, respectively, with a linear fitting slope of 0.0225 MPa / μm per period and a variance of 0.0012. The location of the maximum stress moves from the interface between the upper adhesive layer and the microprism array layer to 5 micrometers inside the microprism array layer. Optical response sensitivity information is extracted by perturbation testing of the optical performance feature set. Key parameters such as the privacy protection threshold angle, color gamut coverage, and color distortion were selected. A small perturbation of ±2% was applied to each parameter, and the changes in the responses of other parameters were observed. The ratio of the change amplitude to the perturbation amplitude was defined as the sensitivity coefficient. For example, when the privacy protection threshold angle increased from 35 degrees to 35.7 degrees, the color gamut coverage decreased from 96% to 94.5%, and the color distortion increased from 1.2 to 1.5. The calculated sensitivity coefficients of the privacy protection threshold angle to color gamut coverage were 3.75, and to color distortion were 7.5. A sensitivity matrix was constructed from the sensitivity coefficients of all parameters.

[0028] The deep neural network encoding layer adopts a multilayer perceptron structure, including an input layer, two hidden layers, and an output layer. The input layer has 10 neurons, receiving a 4-dimensional stress state feature set and a 6-dimensional optical performance feature set. The first hidden layer has 64 neurons, and the second hidden layer has 32 neurons, both using ReLU activation. The output layer has 16 neurons, outputting a 16-dimensional initial co-representation vector. During encoding, the three features of stress evolution trend information are concatenated with the stress state feature set, and the sensitivity matrix of optical response sensitivity information is dimensionality-reduced, and five principal components are extracted and concatenated with the optical performance feature set. After concatenation, the input feature dimension is expanded to 18 dimensions. The network is trained using the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 500 training iterations. The training data comes from a state history database, with a training set to validation set ratio of 8:2.

[0029] The initial collaborative representation vector is mapped to the collaborative state mapping space using principal component analysis for dimensionality reduction. The 16-dimensional vector is projected onto the directions of the first two principal components to obtain two-dimensional coordinate points. The cumulative variance contribution rate of the first two principal components exceeds 85%. These mapped two-dimensional coordinate points are the current state points. For example, the current state point coordinates are a stress standard deviation of 0.18 MPa and an optical comprehensive score of 83. The deviation between the current state point and the collaborative optimal region is calculated in two steps. The ideal collaborative state reference point is defined as the centroid of the candidate points within the collaborative optimal region. In this example, the centroid coordinates are a stress standard deviation of 0.11 MPa and an optical comprehensive score of 91. The Euclidean distance from the current state point to the centroid is 10.63 units, which is taken as the initial deviation. The ray casting method is used to determine that the current state point is outside the boundary of the collaborative optimal region. The shortest vertical distance from the point to the boundary is calculated as 6.8 units, which is taken as the shortest penetration distance. The initial deviation and the penetration distance are weighted and summed, with a weight coefficient of 0.5 for each, resulting in a final deviation of 8.715 units. The collaborative tolerance threshold is set to 5 units. If the current final deviation exceeds the threshold, the decomposition and compensation process is triggered.

[0030] The initial co-representation vector decomposition employed the FastICA independent component analysis algorithm, with an iterative convergence threshold of 0.0001 and a maximum of 200 iterations. The 16-dimensional vector was decomposed into two 8-dimensional independent components. Based on a predefined component assignment rule, dimensions highly correlated with stress state characteristics were assigned to the stress-independent component, and dimensions highly correlated with optical performance characteristics were assigned to the optical-independent component. The final deviation was decomposed into a stress-dimensional deviation component and an optical-dimensional deviation component, with the stress-dimensional deviation component having a value of 0.07 MPa and the optical-dimensional deviation component having a value of 8 points. The contribution ratio of the stress-independent component to the stress-dimensional deviation component was calculated to be 0.65, and the contribution ratio of the optical-independent component to the optical-dimensional deviation component was 0.78. A compensation vector pointing to the ideal reference point projected onto the stress subspace was applied to the stress-independent component, with the projection coordinates being the stress standard deviation of 0.11 MPa, and the compensation amplitude modulated by the contribution ratio of 0.65. Similarly, a compensation vector pointing to the ideal reference point projected onto the optical subspace was applied to the optical-independent component, with the projection coordinates being the optical comprehensive score of 91 points, and the compensation amplitude modulated by the contribution ratio of 0.78. The compensated stress-independent components and optical-independent components are fused by inverse transformation of independent component analysis. Using the mixing matrix during decomposition, the two compensated 8-dimensional independent components are restored to a 16-dimensional vector, which is the co-representation vector.

[0031] This invention constructs a cooperative state mapping space and records historical iterative state trajectories, enabling the determination of the optimal cooperative region based on historical data, thus realizing a data-driven method for optimizing privacy film performance. Through temporal unfolding of the stress state feature set and perturbation testing of the optical performance feature set, more predictive evolutionary trends and response sensitivity information are extracted, enhancing the adaptability to dynamic changes.

[0032] In one optional implementation, the steps of calculating the deviation between the current state point and the optimal cooperative region, and when the deviation exceeds the cooperative tolerance threshold, decomposing the initial cooperative characterization vector into stress-independent components and optical-independent components, applying reverse compensation based on the deviation contribution of each independent component, and then re-fusing to obtain the cooperative characterization vector include: The Euclidean distance from the current state point to the ideal cooperative state reference point of the cooperative optimal region is calculated as the initial deviation. When the current state point is located outside the boundary of the cooperative optimal region, the shortest penetration distance to the boundary is calculated, and the weighted sum of the initial deviation and the penetration distance is used as the final deviation. When the final deviation exceeds the cooperative tolerance threshold, the initial cooperative characterization vector is decomposed into statistically independent stress independent components and optical independent components using the independent component analysis algorithm; the final deviation is decomposed into stress dimension deviation components and optical dimension deviation components, and the contribution ratio of each independent component to the corresponding dimension deviation component is calculated. A compensation vector pointing to the projection of the ideal cooperative state reference point onto the stress subspace is applied to the stress-independent component, and a compensation vector pointing to the projection of the ideal cooperative state reference point onto the optical subspace is applied to the optical-independent component. The compensation amplitude is modulated by the contribution ratio of each independent component. The compensated stress-independent component and optical-independent component are fused to obtain the cooperative characterization vector.

[0033] For example, the ideal cooperative state reference point is defined as the centroid of all candidate points within the optimal cooperative region. The centroid coordinates are obtained by averaging the stress distribution state dimension coordinates of all points in the candidate point set, and by averaging the optical performance state dimension coordinates of all points, resulting in the optical dimension centroid coordinates. In the specific implementation, each candidate point in the candidate point set is traversed, its stress distribution state dimension coordinates are accumulated and divided by the total number of candidate points to obtain the stress dimension centroid coordinates. Similarly, its optical performance state dimension coordinates are accumulated and divided by the total number of candidate points to obtain the optical dimension centroid coordinates. The candidate point set is derived from trajectory points in the state history database that meet the deviation threshold conditions. The deviation threshold conditions are that the stress deviation does not exceed a set stress deviation threshold and the optical performance deviation does not exceed a set optical performance deviation threshold. The stress deviation is the absolute value of the difference between the stress distribution state dimension coordinates of the trajectory point and the target stress distribution state value, and the optical performance deviation is the absolute value of the difference between the optical performance state dimension coordinates of the trajectory point and the target comprehensive optical performance score.

[0034] The Euclidean distance from the current state point to the ideal cooperative state reference point is used as the initial deviation. The Euclidean distance is calculated using a two-dimensional planar distance formula, by adding the square root of the difference between the stress distribution state dimension coordinates of the current state point and the stress dimension centroid coordinates of the ideal cooperative state reference point, and the square root of the difference between the optical performance state dimension coordinates of the current state point and the optical dimension centroid coordinates of the ideal cooperative state reference point. To determine whether the current state point is outside the boundary of the cooperative optimal region, a ray method is used. A ray is emitted from the current state point in any direction, and the number of intersections between the ray and the boundary polygon of the cooperative optimal region is counted. An odd number of intersections indicates the point is inside the polygon, and an even number indicates the point is outside. The boundary polygon of the cooperative optimal region is obtained by calculating the convex hull of the candidate point set using the Graham scan algorithm. The candidate point set is sorted by polar angle, and then the collinearity or turning relationship of three points is determined sequentially, retaining the vertices that constitute the convex hull.

[0035] When the current state point is located outside the boundary of the optimal collaborative region, calculate the shortest penetration distance from the current state point to the boundary of the optimal collaborative region. Traverse all edges of the polygonal boundary of the optimal collaborative region, and calculate the perpendicular distance from the current state point to each edge. The perpendicular distance is calculated using the point-to-line distance formula. Substitute the coordinates of the two endpoints of the edge into the equation of the line to obtain the general coefficients of the line, and then substitute the coordinates of the current state point into the point-to-line distance formula to obtain the perpendicular distance of the edge. Take the minimum perpendicular distance among all edges as the shortest penetration distance. In the data example, the polygonal boundary of the optimal collaborative region consists of the convex hull of 37 candidate points and has 12 edges. Traverse these 12 edges to calculate the perpendicular distance from the current state point to each edge; the minimum perpendicular distance is 6.8 units, which is taken as the shortest penetration distance. The initial deviation and the penetration distance are weighted and summed to obtain the final deviation. The weight coefficients for the weighted sum are all set to 0.5. The calculation formula is: final deviation equals initial deviation multiplied by a weight coefficient of 0.5 plus penetration distance multiplied by a weight coefficient of 0.5.

[0036] When the final deviation exceeds the collaborative tolerance threshold, the decomposition and compensation process is triggered. The collaborative tolerance threshold is set to 5 units. The initial collaborative representation vector is decomposed into statistically independent stress and optical components using the FastICA independent component analysis algorithm. The FastICA algorithm's iteration convergence threshold is set to 0.0001, and the maximum number of iterations is set to 200. The initial collaborative representation vector is a 16-dimensional vector, decomposed into two 8-dimensional independent components. Component assignment rules are determined by calculating the correlation coefficients between each dimension element of the initial collaborative representation vector and the stress state feature set and the optical performance feature set. The Pearson correlation coefficient is used. The 8 dimensions with higher correlation coefficients to the stress state feature set are assigned to the stress independent components, and the 8 dimensions with higher correlation coefficients to the optical performance feature set are assigned to the optical independent components. The stress state feature set contains 4 feature dimensions, and the optical performance feature set contains 6 feature dimensions. When calculating the correlation coefficients, the correlation coefficients between each dimension element and each feature dimension of the corresponding feature set are calculated separately, and then the average value is taken.

[0037] The final deviation is decomposed into stress dimension deviation components and optical dimension deviation components. The stress dimension deviation component is the absolute value of the difference between the stress distribution state dimension coordinates of the current state point and the stress dimension centroid coordinates of the ideal coherent state reference point. The optical dimension deviation component is the absolute value of the difference between the optical performance state dimension coordinates of the current state point and the optical dimension centroid coordinates of the ideal coherent state reference point. The contribution ratio of the independent stress components to the stress dimension deviation component is obtained by weighted summation of the correlation coefficients between each element of the independent stress component and the stress state feature set, with the weights being the normalized values ​​of the absolute values ​​of each element. The contribution ratio of the independent optical components to the optical dimension deviation component is also obtained by weighted summation of the correlation coefficients between each element of the independent optical component and the optical performance feature set.

[0038] A compensation vector is applied to the stress-independent components, pointing towards the projection of the ideal cooperative state reference point onto the stress subspace. The projection coordinates of the stress subspace are the mappings of the stress dimension centroid coordinates of the ideal cooperative state reference point onto the stress-independent component space. This mapping is obtained by multiplying the stress dimension centroid coordinates of the ideal cooperative state reference point with the mixing matrix of the stress-independent components. The direction of the compensation vector is from the current stress-independent component to the projection coordinates of the stress subspace. The compensation amplitude is modulated by the contribution ratio of the stress-independent component to the stress dimension deviation component, calculated as the magnitude of the compensation vector multiplied by the contribution ratio of 0.65. Similarly, a compensation vector is applied to the optical-independent components, pointing towards the projection of the ideal cooperative state reference point onto the optical subspace. The projection coordinates of the optical subspace are the mappings of the optical dimension centroid coordinates of the ideal cooperative state reference point onto the optical-independent component space. The direction of the compensation vector is from the current optical-independent component to the projection coordinates of the optical subspace. The compensation amplitude is modulated by the contribution ratio of the optical-independent component to the optical dimension deviation component of 0.78. The compensated stress independent components and optical independent components are fused to obtain a co-representation vector. The fusion adopts the inverse transformation of independent component analysis. Using the mixing matrix obtained during decomposition, the two 8-dimensional independent components after compensation are concatenated into a 16-dimensional vector and then multiplied with the inverse matrix of the mixing matrix to restore the 16-dimensional co-representation vector.

[0039] This invention employs a weighted combination of Euclidean distance and penetration distance to calculate deviation, providing a more accurate state assessment mechanism. By introducing an independent component analysis algorithm to decompose the collaborative characterization vector, it can accurately identify the independent influencing factors of stress and optics, achieving precise compensation for different dimensions. This makes the collaborative characterization vector closer to the ideal state, significantly improving the collaborative optimization effect.

[0040] In one optional implementation, the step of identifying the interaction dependency between stress state characteristics and optical performance characteristics through an attention mechanism and establishing a bidirectional influence weight matrix includes: The stress state feature set is converted into a query vector, and the optical performance feature set is converted into a key vector and a value vector. The correlation strength is calculated by the inner product operation of the query vector and the key vector. The correlation strength is used as a weight to perform weighted aggregation on the value vector to obtain a feature representation containing interactive dependencies, and an interactive dependency matrix is ​​constructed. Identify the positive influence path of stress state characteristics on optical performance characteristics from the interaction dependency matrix. The positive influence path represents the transmission relationship from stress evolution to optical performance change. Calculate the transmission weight coefficient of each feature node in the positive influence path. Identify the reverse constraint path of optical performance characteristics on stress state characteristics from the interaction dependency matrix. The reverse constraint path represents the feedback relationship between optical optimization requirements and stress distribution adjustment. Calculate the feedback weight coefficient of each feature node in the reverse constraint path. A two-way influence weight matrix is ​​constructed based on the transmitted weight coefficient and the feedback weight coefficient.

[0041] For example, the stress state feature set contains four feature dimensions: stress peak location coordinates, stress gradient value, stress distribution uniformity standard deviation, and maximum stress amplitude. The query vector transformation uses a linear transformation matrix with 64 rows and 4 columns. The matrix elements are randomly initialized and optimized using the training dataset. The stress state feature set is represented as a 4-dimensional column vector, which is multiplied by the linear transformation matrix to obtain a 64-dimensional query vector. The linear transformation matrix is ​​trained using a gradient descent algorithm with a learning rate of 0.001, a batch size of 32, and 1000 training iterations. The training data comes from paired samples of the stress state feature set and optical performance feature set in the state history database.

[0042] The optical performance feature set comprises six dimensions: color gamut coverage, privacy threshold angle, color distortion, brightness uniformity, viewing angle dependence, and transmittance. Key vector transformation employs the same linear transformation mechanism as the query vector, using an independent key vector linear transformation matrix with dimensions of 64 rows and 6 columns. The matrix elements are optimized through the same training process as the query vector linear transformation matrix. The optical performance feature set is represented as a 6-dimensional column vector, multiplied by the key vector linear transformation matrix to obtain a 64-dimensional key vector. Value vector transformation also uses an independent value vector linear transformation matrix with dimensions of 64 rows and 6 columns. The optical performance feature set is multiplied by the value vector linear transformation matrix to obtain a 64-dimensional value vector.

[0043] When calculating the association strength using the inner product of the query vector and the key vector, the corresponding elements of the 64-dimensional query vector and the 64-dimensional key vector are multiplied one by one and then summed to obtain a scalar association strength value. The result of the inner product operation needs to be normalized by a scaling factor set to the square root of the query vector dimension, i.e., 8. Dividing the inner product result by the scaling factor 8 yields the normalized association strength. The normalized association strength is then subjected to a probabilistic transformation using the Softmax function. The Softmax function calculates the association strength by taking the exponent of the association strength and dividing it by the sum of all association strength exponent values. In the actual implementation, each feature dimension in the stress state feature set independently generates a query vector, and each feature dimension in the optical performance feature set independently generates a key vector, forming a 4x6 query-key pair combination. Each combination calculates an association strength value, forming a 4x6 association strength matrix.

[0044] When using correlation strength as a weight to perform weighted aggregation of value vectors, the value vector corresponding to each feature dimension in the optical performance feature set is weighted and summed with all correlation strength values ​​in the corresponding column of the correlation strength matrix for that feature dimension. The weighted summation is calculated by multiplying each correlation strength value by one element of the corresponding value vector and then summing the results to obtain the aggregated value vector. Since there are four stress state feature dimensions, each corresponding to a row of correlation strength values, each stress state feature dimension generates an aggregated 64-dimensional value vector. The four stress state feature dimensions generate four aggregated value vectors, which are then stacked row-wise to form a 4x64 feature representation matrix. This feature representation matrix contains the interactive dependencies between stress state features and optical performance features, with each row representing the projection of a stress state feature dimension into the optical performance feature space.

[0045] When constructing the interaction dependency matrix, the feature representation matrix is ​​concatenated with the value vector matrix of the original optical performance feature set. The original optical performance feature set's value vector matrix is ​​6 rows and 64 columns, composed of stacked 64-dimensional value vectors corresponding to the six optical performance feature dimensions. The concatenation operation merges the 4-row, 64-column stress feature representation matrix with the 6-row, 64-column optical value vector matrix along the row direction, resulting in a 10-row, 64-column interaction dependency matrix. The first four rows of the interaction dependency matrix represent the stress state features under optical influence, and the last six rows represent the original representation of the optical performance features. To enhance the matrix's symmetry and bidirectional correlation capability, the interaction dependency matrix is ​​transposed and then multiplied again with the original matrix, resulting in a 10-row, 10-column symmetric interaction dependency matrix. The row and column indices of the symmetric interaction dependency matrix correspond to the stress state feature dimension and the optical performance feature dimension, respectively, and the matrix element values ​​represent the interaction dependency strength between the corresponding feature dimensions.

[0046] When identifying the positive influence path of stress state features on optical performance features from the interaction dependency matrix, the positive influence path is defined as the set of directed edges pointing from nodes in the stress state feature dimension to nodes in the optical performance feature dimension. The first four rows of the symmetric interaction dependency matrix correspond to the stress state feature dimension, and the last six columns correspond to the optical performance feature dimension. The submatrix at the intersection of the first four rows and the last six columns is extracted as a candidate matrix for the positive influence path. The candidate matrix has a dimension of 4 rows and 6 columns. A threshold filter is applied to each element of the candidate matrix. The threshold is set to the median of the absolute values ​​of all elements in the candidate matrix. Elements with absolute values ​​below the threshold are set to zero, while elements with absolute values ​​above or equal to the threshold retain their original values. The retained elements form the edges of the positive influence path. The starting point of each edge is the stress state feature dimension corresponding to the row index, and the ending point is the optical performance feature dimension corresponding to the column index. The weight of the edge is the value of the matrix element.

[0047] The transfer weight coefficient reflects the importance of a node in the positive influence path. For each feature node, the number of times it serves as the start or end point of a positive influence path edge is counted; the number of times it serves as the start point is called the out-degree, and the number of times it serves as the end point is called the in-degree. The transfer weight coefficient of a node is calculated by dividing the sum of its out-degree and in-degree by the normalized value of the sum of the out-degree and in-degree of all nodes. Stress state feature nodes only serve as the start point of a positive influence path edge, and their transfer weight coefficient is determined by their out-degree. Optical performance feature nodes only serve as the end point of a positive influence path edge, and their transfer weight coefficient is determined by their in-degree. For example, the stress peak position coordinate node is used as the starting point to connect four positive influence path edges with an out-degree of 4 and a transfer weight coefficient of 0.27; the stress gradient value node has an out-degree of 3 and a transfer weight coefficient of 0.20; the stress distribution uniformity standard deviation node has an out-degree of 5 and a transfer weight coefficient of 0.33; the maximum stress amplitude node has an out-degree of 3 and a transfer weight coefficient of 0.20; the color gamut coverage node has an in-degree of 3 and a transfer weight coefficient of 0.20; the privacy protection critical angle node has an in-degree of 2 and a transfer weight coefficient of 0.13; and the brightness uniformity node has an in-degree of 4 and a transfer weight coefficient of 0.27.

[0048] The reverse constraint path is defined as the set of directed edges pointing from nodes in the optical performance feature dimension to nodes in the stress state feature dimension. The last six rows of the symmetric interaction dependency matrix correspond to the optical performance feature dimension, and the first four columns correspond to the stress state feature dimension. The submatrix at the intersection of the last six rows and the first four columns is extracted as the candidate matrix for the reverse constraint path. The candidate matrix has a dimension of 6 rows and 4 columns. The same threshold filtering mechanism as for the forward influence path is applied to the candidate matrix. The threshold is set to the median of the absolute values ​​of all elements in the candidate matrix. The elements retained after filtering constitute the edges of the reverse constraint path. For example, if the median of the absolute values ​​of all elements in the candidate matrix is ​​8.9, the element value of 17.3 for color gamut coverage and stress peak position coordinates is higher than the threshold and is retained; the element value of -6.8 for the anti-spying critical angle and stress gradient value is lower than the threshold and is set to zero; and the element value of 11.6 for the standard deviation of brightness uniformity and stress distribution uniformity is higher than the threshold and is retained, a total of 13 reverse constraint path edges are retained after filtering.

[0049] The feedback weight coefficient reflects the importance of the node in the reverse constraint path. For each feature node, the number of times it serves as the start or end point of a reverse constraint path edge is counted. Optical performance feature nodes only serve as the start point of a reverse constraint path edge, and their feedback weight coefficient is determined by their out-degree. Stress state feature nodes only serve as the end point of a reverse constraint path edge, and their feedback weight coefficient is determined by their in-degree. The feedback weight coefficient of a node is calculated as the normalized value of the node's out-degree or in-degree divided by the sum of the out-degrees and in-degrees of all nodes. For example, the color gamut coverage node serves as the starting point, connecting three reverse constraint path edges with an out-degree of 3 and a feedback weight coefficient of 0.23; the anti-peeping critical angle node has an out-degree of 2 and a feedback weight coefficient of 0.15; the color distortion node has an out-degree of 1 and a feedback weight coefficient of 0.08; the brightness uniformity node has an out-degree of 4 and a feedback weight coefficient of 0.31; the viewing angle dependence node has an out-degree of 2 and a feedback weight coefficient of 0.15; the transmittance node has an out-degree of 1 and a feedback weight coefficient of 0.08; the stress peak position coordinate node has an in-degree of 4 and a feedback weight coefficient of 0.31; the stress gradient value node has an in-degree of 3 and a feedback weight coefficient of 0.23; the stress distribution uniformity standard deviation node has an in-degree of 5 and a feedback weight coefficient of 0.38; and the maximum stress amplitude node has an in-degree of 1 and a feedback weight coefficient of 0.08.

[0050] The bidirectional influence weight matrix has a dimension of 10 rows and 10 columns, with row and column indices corresponding to four stress state feature dimensions and six optical performance feature dimensions, respectively. The first four rows and last six columns of the matrix are filled with the edge weights of the positive influence path, with each element value being the weight of the edge in the positive influence path multiplied by the geometric mean of the transfer weight coefficients of the starting and ending nodes. The last six rows and first four columns of the matrix are filled with the edge weights of the negative constraint path, with each element value being the weight of the edge in the negative constraint path multiplied by the geometric mean of the feedback weight coefficients of the starting and ending nodes. The first four rows and first four columns of the matrix and the last six rows and last six columns are set as a diagonal matrix, with diagonal elements representing the transfer or feedback weight coefficients of the corresponding nodes, and off-diagonal elements having a value of zero.

[0051] This invention achieves accurate calculation of the correlation between features and in-depth mining of dependencies through an attention mechanism. It can simultaneously identify positive influence paths and negative constraint paths, construct a bidirectional influence weight matrix, and comprehensively quantify the bidirectional relationship of stress-optics interaction, providing a more comprehensive basis for strategy generation.

[0052] In one optional implementation, the step of inputting the collaborative representation vector into a policy generation network, wherein the policy generation network analyzes the positive influence path of stress modulation on optical performance and the negative constraint path of optical optimization on stress distribution based on the bidirectional influence weight matrix, and generates a collaborative control policy including stress modulation instructions and optical compensation instructions, includes: The collaborative representation vector is input into the policy generation network. Based on the stress evolution trend information and optical response sensitivity information embedded in the collaborative representation vector, the stress distribution deviation vector and optical performance deviation vector between the current state and the target state are analyzed. Based on the transfer weight coefficients in the bidirectional influence weight matrix, the influence contribution of the stress distribution deviation vector on each optical performance feature is traced to construct a path diagram of the effect of stress regulation on optical performance; based on the feedback weight coefficients in the bidirectional influence weight matrix, the constraint strength of the optical performance deviation vector on each stress state feature is traced to construct a path diagram of the reverse constraint of optical optimization on stress distribution. Stress control nodes whose influence contribution exceeds the contribution threshold are extracted from the action path diagram, and stress control instructions are generated according to the weight distribution of the stress control nodes; optical compensation nodes whose constraint strength exceeds the constraint threshold are extracted from the reverse constraint path diagram, and optical compensation instructions are generated according to the weight distribution of the optical compensation nodes. Conflict detection is performed on stress control commands and optical compensation commands. When antagonistic influences are detected, priority ranking and coordination correction are performed based on the path weights of the action path diagram and the reverse constraint path diagram to obtain a cooperative control strategy.

[0053] Combination Figure 2 The flowchart illustrating the collaborative control strategy generation and conflict resolution process for high color gamut privacy films is provided. The strategy generation network is implemented using a multilayer perceptron architecture, comprising an input layer, three hidden layers, and an output layer. The input layer receives a collaborative representation vector, which has 16 dimensions. The first 8 dimensions contain stress evolution trend information, and the last 8 dimensions contain optical response sensitivity information. The number of neurons in the hidden layer is 256, 128, and 64, respectively, and the ReLU function is used as the activation function, which sets negative values ​​to zero and retains positive values. The output layer has 10 neurons, corresponding to four stress state feature dimensions and six optical performance feature dimensions. The output layer does not use an activation function and directly outputs linear values. The strategy generation network is trained using the backpropagation algorithm, with the mean squared error loss function, a learning rate of 0.0005, and a batch size of 16. The training data comes from a labeled state deviation sample and control command pairing dataset, and the training iterations are performed for 2000 rounds until the validation set loss converges.

[0054] When analyzing the stress distribution deviation vector and optical performance deviation vector between the current state and the target state based on the stress evolution trend information and optical response sensitivity information embedded in the collaborative representation vector, the first four elements of the output layer of the strategy generation network correspond to the four components of the stress distribution deviation vector, and the last six elements correspond to the six components of the optical performance deviation vector. The four components of the stress distribution deviation vector represent the deviation of the stress peak position coordinates, the deviation of the stress gradient value, the deviation of the stress distribution uniformity standard deviation, and the deviation of the maximum stress amplitude, respectively. The deviation value is the difference between the target state and the current state. The six components of the optical performance deviation vector represent the deviation of color gamut coverage, the deviation of the privacy threshold angle, the deviation of color distortion, the deviation of brightness uniformity, the deviation of viewing angle dependence, and the deviation of transmittance, respectively. The sign of the deviation vector indicates the adjustment direction; a positive value indicates that the feature value needs to be increased, and a negative value indicates that the feature value needs to be decreased. The absolute value represents the adjustment magnitude. To enhance the stability of deviation analysis, a dynamic range limit is imposed on each component of the deviation vector. The limit range is determined by the 95th percentile and 5th percentile of the deviation components in the training dataset. Deviation values ​​exceeding the range are truncated to the boundary value.

[0055] When tracing the contribution of the stress distribution deviation vector to various optical performance characteristics based on the transferred weight coefficients in the two-way influence weight matrix, the contribution is calculated using matrix-vector multiplication. The first four rows and last six columns of the two-way influence weight matrix are extracted as the positive influence matrix, which has a dimension of 4 rows and 6 columns, and its elements are the influence weights of stress state characteristics on optical performance characteristics. The stress distribution deviation vector is represented as a 4-dimensional column vector, and matrix multiplication is performed with the positive influence matrix to obtain a 6-dimensional influence contribution vector. Each component of the influence contribution vector represents the contribution of the corresponding optical performance characteristic to the influence of stress regulation; the sign of the contribution indicates the direction of influence, and the absolute value indicates the intensity of influence. To eliminate the influence of dimensions, the influence contribution vector is normalized by dividing each component by the sum of the absolute values ​​of all components. The sum of the absolute values ​​of the components of the normalized influence contribution vector is 1.

[0056] The path graph of stress modulation's effect on optical performance is represented using a directed graph data structure. The graph contains four stress state feature nodes and six optical performance feature nodes, with directed edges pointing from stress state feature nodes to optical performance feature nodes. The edge weight is determined by the product of the corresponding element value of the positive influence matrix and the corresponding component of the stress distribution deviation vector; the product represents the actual influence intensity of the edge under the current deviation state. Edges with absolute weight values ​​below the minimum weight threshold are deleted from the graph. The minimum weight threshold is set to the 10th percentile of the absolute weight values ​​of all edges, reducing the complexity of the path graph. The retained edges are arranged in descending order of absolute weight value, forming the edge set of the path graph. The path graph is stored as an adjacency list structure, where each stress state feature node records its outgoing edge list, which includes the index of the target optical performance feature node and the edge weight.

[0057] When tracing the constraint strength of the optical performance deviation vector on each stress state feature based on the feedback weight coefficients in the two-way influence weight matrix, the constraint strength calculation also employs matrix-vector multiplication. The last six rows and first four columns of the two-way influence weight matrix are extracted as the reverse constraint matrix. The reverse constraint matrix has a dimension of 6 rows and 4 columns, and its elements are the constraint weights of the optical performance features on the stress state features. The optical performance deviation vector is represented as a 6-dimensional column vector, and matrix multiplication is performed with the reverse constraint matrix to obtain a 4-dimensional constraint strength vector. Each component of the constraint strength vector represents the intensity of the corresponding stress state feature under optical optimization constraint. The constraint strength vector is also normalized to ensure that the sum of the absolute values ​​of each component is 1.

[0058] The reverse constraint path graph for stress distribution in optical optimization is also represented using a directed graph data structure, with directed edges pointing from optical performance feature nodes to stress state feature nodes. The edge weight is determined by the product of the corresponding element value of the reverse constraint matrix and the corresponding component of the optical performance deviation vector. Edges with absolute weight values ​​below the minimum weight threshold are deleted from the graph. The minimum weight threshold is set to the 10th percentile of the absolute values ​​of all edge weights. The reverse constraint path graph is stored as an adjacency list structure, with each optical performance feature node recording its outgoing edge list.

[0059] In the action path graph, the contribution threshold is set as the mean of the absolute values ​​of the normalized influence contribution vector components. This mean is calculated by summing the absolute values ​​of the six components of the influence contribution vector and dividing by 6. All edges in the action path graph are traversed, and the influence contribution of each edge's target optical performance feature node is checked to see if it exceeds the contribution threshold. Edges exceeding the threshold have their starting stress state feature nodes marked as stress control nodes. The weight distribution of a stress control node is determined by the sum of the absolute values ​​of the weights of all outgoing edges from that node. Weight distribution normalization ensures that the sum of the weights of all stress control nodes is 1.

[0060] The stress control command includes three fields: control target, control direction, and control amplitude. The control target field identifies the type of stress state characteristic to be controlled, and its value is the name of the characteristic dimension corresponding to the stress control node. The control direction field identifies the direction of the control operation; its value is determined by the sign of the corresponding component of the stress distribution deviation vector, with positive values ​​corresponding to the increasing direction and negative values ​​to the decreasing direction. The control amplitude field identifies the magnitude of the control operation; its value is calculated by multiplying the absolute value of the corresponding component of the stress distribution deviation vector by the normalized weight of the stress control node. The product is saturated to prevent excessively large single control amplitudes; the upper bound of the saturation limit is set to 20% of the normal fluctuation range of the characteristic dimension. The generated stress control commands are arranged in descending order of control amplitude, with commands having larger amplitudes executed first.

[0061] The constraint threshold is set to the mean of the absolute values ​​of all components of the normalized constraint strength vector. All edges of the reverse constraint path graph are traversed, and the constraint strength of the target stress state feature node of each edge is checked to see if it exceeds the constraint threshold. The starting optical performance feature node of the edge exceeding the threshold is marked as an optical compensation node. The weight distribution of the optical compensation node is determined by the sum of the absolute values ​​of the edge weights of all outgoing edges of that node. The weight distribution is normalized so that the sum of the weights of all optical compensation nodes is 1.

[0062] The optical compensation command includes a compensation target field, a compensation direction field, and a compensation magnitude field. The compensation target field identifies the type of optical performance feature to be compensated, and its value is the name of the feature dimension corresponding to the optical compensation node. The compensation direction field is determined by the sign of the corresponding component of the optical performance deviation vector. The compensation magnitude field is calculated by multiplying the absolute value of the corresponding component of the optical performance deviation vector by the normalized weight of the optical compensation node. The product is also subject to saturation constraints, with the upper bound set to 15% of the normal fluctuation range of the feature dimension. The generated optical compensation commands are arranged in descending order of compensation magnitude.

[0063] Conflict detection analysis checks whether two sets of instructions produce opposing adjustment requirements for the same physical parameter. A physical parameter mapping table is established, recording the underlying physical parameter set corresponding to each stress state characteristic and optical performance characteristic. Physical parameters include substrate thickness, stress layer thickness, material refractive index, and interface stress distribution. The stress control instruction list and optical compensation instruction list are traversed, extracting the physical parameter set corresponding to each instruction, and detecting whether the same physical parameter is required to be adjusted in opposite directions in both sets of instructions. When the same physical parameter is required to increase in the stress control instruction but decreased in the optical compensation instruction, or vice versa, it is determined to be an antagonistic influence. All detected antagonistic influences and their corresponding stress control instruction and optical compensation instruction pairs are recorded. For example, a stress peak position coordinate control instruction requires an increase in substrate thickness of 0.8 micrometers, while a color gamut coverage compensation instruction requires a decrease in substrate thickness of 0.3 micrometers. An antagonistic influence on the substrate thickness parameter is detected, and this pair is recorded as a conflict instruction pair.

[0064] When an adversarial influence is detected, the path weight is defined as the product of the absolute values ​​of all edge weights from the starting node to the conflicting physical parameter node. For stress control commands, the path weight is calculated by traversing the action path graph from the corresponding stress control node to the optical performance feature node, and then from the optical performance feature node to the conflicting physical parameter node along the physical parameter mapping relationship. The product of the absolute values ​​of all edge weights on the path is used as the stress control path weight. For optical compensation commands, the path weight is calculated by traversing the reverse constraint path graph from the corresponding optical compensation node to the stress state feature node, and then from the physical parameter mapping relationship to the conflicting physical parameter node. The reverse constraint path weight is calculated. The stress control path weight and the reverse constraint path weight corresponding to the same conflicting physical parameter are compared, and commands with larger weights have higher priority. Commands with higher priority maintain their original control direction and amplitude, while commands with lower priority are coordinated and corrected. The correction method is to reduce the control amplitude of the command to be equal to that of the command with higher priority and then take the opposite sign. The corrected control amplitude must not exceed 50% of the original control amplitude; if it exceeds 50%, the command is directly canceled.

[0065] The collaborative control strategy includes fields for execution order, instruction type, control target, control direction, and control amplitude. The execution order field sorts instructions according to their priority, with higher-priority instructions executed first. The instruction type field identifies whether the instruction belongs to the stress control or optical compensation type. The collaborative control strategy is output in a structured data format, supporting direct parsing and execution by the equipment control module.

[0066] The strategy generation network of this invention can analyze the deviation vector based on the collaborative representation vector and construct the action path graph and the reverse constraint path graph using a bidirectional influence weight matrix, thereby realizing collaborative decision-making for stress control and optical compensation. This method can extract key control nodes and generate precise instructions based on the weight distribution, significantly improving the efficiency and accuracy of strategy formulation in the privacy screen protector optimization process.

[0067] In one optional implementation, conflict detection is performed on the stress control command and the optical compensation command. When an antagonistic influence is detected, the steps of prioritizing and coordinating correction based on the path weights of the action path diagram and the reverse constraint path diagram include: Construct an influence matrix for stress control commands and optical compensation commands, where each element represents the mutual influence coefficient between corresponding command pairs; identify command pairs with negative mutual influence coefficients from the influence matrix and mark them as conflicting command pairs; For each conflicting command pair, extract the path containing the stress control command in the conflicting command pair from the action path diagram, and calculate the cumulative path weight of the path; extract the path containing the optical compensation command in the conflicting command pair from the reverse constraint path diagram, and calculate the cumulative path weight of the path; compare the cumulative path weights of the two paths, and set the command with the larger cumulative path weight as the high priority command. The lower priority instruction in the conflicting instruction pair is attenuated by a factor that is determined by the ratio of the cumulative path weight of the higher priority instruction to the cumulative path weight of the lower priority instruction.

[0068] For example, an influence matrix is ​​constructed to quantify the interaction between instructions. The rows of the influence matrix correspond to all stress control instructions, and the columns correspond to all optical compensation instructions. The matrix dimension is determined by the number of instructions. The elements of the influence matrix are calculated by analyzing the overlap of physical parameters affected by the instructions. A physical parameter mapping table is established to record the set of underlying physical parameters corresponding to each stress state characteristic and optical performance characteristic. These physical parameters include the thickness of each functional layer, interlayer stress value, material refractive index, microprism geometric parameters, and interface bonding strength. For any stress control instruction, the set of physical parameters corresponding to its control target is extracted as the stress parameter set; for any optical compensation instruction, the set of physical parameters corresponding to its compensation target is extracted as the optical parameter set. The intersection of the two parameter sets is calculated to obtain a list of commonly affected physical parameters. For each commonly affected physical parameter, the adjustment direction required by the two instructions is checked. The adjustment direction of the stress control instruction is determined by the sign of the corresponding component of the stress distribution deviation vector; the adjustment direction of the optical compensation instruction is determined by the sign of the corresponding component of the optical performance deviation vector. When the adjustment directions are the same, the directional consistency coefficient of the physical parameter is set to positive 1; when the adjustment directions are opposite, it is set to negative 1; and the directional consistency coefficient of physical parameters that are not commonly affected is set to 0. The mutual influence coefficient is obtained by weighted summing of the directional consistency coefficients of all the physical parameters that jointly influence each other. The weighting coefficient is determined by the sensitivity of the physical parameter to the system performance; the physical parameter with higher sensitivity has a larger weight. The sensitivity is obtained through statistical analysis of the performance impact of the change of the physical parameter in historical control data. The element in the i-th row and j-th column of the influence matrix is ​​the mutual influence coefficient between the i-th stress control command and the j-th optical compensation command. The coefficient ranges from -1 to +1, and the absolute value represents the intensity of the influence. The sign indicates whether the relationship is cooperative or antagonistic.

[0069] For example, the stress control command includes two instructions: increasing the stress peak position coordinate by 1.41 micrometers and increasing the stress distribution uniformity standard deviation by 0.34 MPa. The optical compensation command includes two instructions: increasing the color gamut coverage by 0.51 percentage points and decreasing the brightness uniformity by 0.11 percentage points. The physical parameter set corresponding to the stress peak position coordinate control command includes the thickness of the upper adhesive layer, the thickness of the microprism array layer, and the interlayer stress distribution. The physical parameter set corresponding to the color gamut coverage compensation command includes the thickness of the microprism array layer, the material refractive index, and the prism angle. The intersection of the two sets is the thickness of the microprism array layer. The stress control command requires increasing the stress peak position coordinate, which corresponds to increasing the thickness of the microprism array layer. The color gamut coverage compensation command requires increasing the color gamut coverage, which corresponds to decreasing the thickness of the microprism array layer to enhance the optical focusing effect. Since the adjustment directions are opposite, the directional consistency coefficient of the microprism array layer thickness is negative 1. The sensitivity weight of the microprism array layer thickness is set to 0.75, and the calculated mutual influence coefficient of this command pair is negative 0.75. The set of physical parameters corresponding to the stress distribution uniformity standard deviation adjustment command includes the thickness of the lower adhesive layer and the thickness of the underlying substrate. These physical parameters have no common influence with the color gamut coverage compensation command, and their mutual influence coefficient is 0. The stress peak position coordinate adjustment command and the brightness uniformity compensation command jointly affect the thickness of the upper adhesive layer. Stress adjustment requires increasing the thickness, while brightness compensation requires decreasing the thickness to reduce interface scattering. The directional consistency coefficient is -1, the upper adhesive layer thickness sensitivity weight is 0.62, and the mutual influence coefficient is -0.62. The stress distribution uniformity standard deviation adjustment command and the brightness uniformity compensation command jointly affect the thickness of the underlying substrate. Both are adjusted in the same direction, and their mutual influence coefficient is positive 0.53. The constructed influence matrix is ​​2 rows and 2 columns. The element in the first row and first column is -0.75, the element in the first row and second column is -0.62, the element in the second row and first column is 0, and the element in the second row and second column is positive 0.53.

[0070] When identifying the mutual influence coefficient, all elements of the influence matrix are traversed, and matrix elements with values ​​less than 0 are selected. The stress control command corresponding to the row index of this element and the optical compensation command corresponding to the column index form a conflict command pair. The conflict command pair is recorded as a triple containing the stress control command identifier, the optical compensation command identifier, and the mutual influence coefficient, and stored in the conflict list. The conflict list is sorted in descending order of the absolute value of the mutual influence coefficient, and conflict pairs with larger absolute values ​​are processed first.

[0071] For each conflict command pair, when extracting the path containing the stress control command from the action path graph, the action path graph is a directed graph structure. Nodes include stress state characteristic nodes and optical performance characteristic nodes. Directed edges point from stress state characteristic nodes to optical performance characteristic nodes, and edge weights are determined by the product of the positive influence matrix element value and the stress distribution deviation vector component. A stress control command corresponds to a stress state characteristic node as the starting node, and the optical performance characteristic node involved in the optical compensation command in the conflict command pair is the target node. The directed path from the starting node to the target node is the path of that stress control command in the action path graph. Path extraction uses a depth-first search algorithm, traversing along the outgoing edges from the starting node, recording the sequence of nodes and edges traversed. When the target node is reached, one path is extracted. If multiple paths exist, the path with the largest sum of absolute edge weights is extracted as the dominant path. The cumulative path weight is defined as the product of the absolute values ​​of all edge weights in the path, and the product represents the overall influence strength of the path. In the data example, the stress control instruction for the first conflict pair is a stress peak position coordinate control instruction, with the corresponding stress peak position coordinate node as the starting node. The optical compensation instruction is a color gamut coverage compensation instruction, involving the color gamut coverage node as the target node. The path from the stress peak position coordinate node to the color gamut coverage node in the action path graph contains one edge with an absolute edge weight of 10.01, and the cumulative path weight is 10.01. The stress control instruction for the second conflict pair is the same, the optical compensation instruction is a brightness uniformity compensation instruction, the target node is the brightness uniformity node, and the path contains an edge from the stress peak position coordinate node to the brightness uniformity node, with an absolute edge weight of 8.34 and a cumulative path weight of 8.34.

[0072] The reverse constraint path graph is a directed graph structure, with directed edges pointing from optical performance feature nodes to stress state feature nodes. Edge weights are determined by the product of the reverse constraint matrix element values ​​and the optical performance deviation vector components. An optical compensation instruction corresponds to an optical performance feature node as the starting node, and the stress state feature node involved in the stress control instruction in the conflict instruction pair serves as the target node. Path extraction also employs a depth-first search algorithm, and the cumulative path weight is the product of the absolute values ​​of all edge weights in the path. In the data example, the first conflict pair's optical compensation instruction is a color gamut coverage compensation instruction, corresponding to the color gamut coverage node as the starting node, and the stress control instruction is a stress peak position coordinate control instruction, involving the stress peak position coordinate node as the target node. The path from the color gamut coverage node to the stress peak position coordinate node in the reverse constraint path graph contains one edge with an absolute edge weight of 4.02, and the cumulative path weight is 4.02. The second conflict pair's optical compensation instruction is a brightness uniformity compensation instruction, with the brightness uniformity node as the starting node and the stress peak position coordinate node as the target node. The absolute value of the path edge weight is 3.58, and the cumulative path weight is 3.58.

[0073] The cumulative path weights of stress control commands in the action path diagram and optical compensation commands in the reverse constraint path diagram are numerically compared. Commands corresponding to paths with larger values ​​have stronger influence transmission capabilities and are set as high-priority commands, while commands corresponding to paths with smaller values ​​are set as low-priority commands. For example, in the first conflict pair, the cumulative path weight of the stress control command is 10.01, and the cumulative path weight of the optical compensation command is 4.02. Since 10.01 is greater than 4.02, the stress peak position coordinate control command is a high-priority command, and the color gamut coverage compensation command is a low-priority command. In the second conflict pair, the cumulative path weight of the stress control command is 8.34, and the cumulative path weight of the optical compensation command is 3.58. Since 8.34 is greater than 3.58, the stress peak position coordinate control command is a high-priority command, and the brightness uniformity compensation command is a low-priority command.

[0074] The lower-priority instruction in a conflicting instruction pair undergoes amplitude attenuation. The attenuation coefficient is determined by the ratio of the cumulative path weight of the higher-priority instruction to the cumulative path weight of the lower-priority instruction. The attenuated amplitude value is obtained by multiplying the adjustment or compensation amplitude of the lower-priority instruction by the attenuation coefficient. This attenuated amplitude value replaces the original amplitude value and updates the adjustment or compensation amplitude field of the instruction. The attenuation coefficient ranges from 0 to 1. A coefficient close to 0 indicates strong suppression of the lower-priority instruction, while a coefficient close to 1 indicates mild suppression. To prevent excessive attenuation from causing instruction failure, a lower limit of 0.2 is set for the attenuation coefficient. When the calculated attenuation coefficient is lower than 0.2, it is forcibly set to 0.2. For example, if the lower-priority instruction in the first conflicting instruction pair is a color gamut coverage compensation instruction with an original compensation amplitude of 0.51%, an attenuation coefficient of 4.02 divided by 10.01 equals 0.40, and the attenuated compensation amplitude is 0.51 multiplied by 0.40 equals 0.20%. The second conflict involves a low-priority instruction for brightness uniformity compensation. The original compensation amplitude is 0.11, and the attenuation coefficient is 3.58 divided by 8.34, which equals 0.43. After attenuation, the compensation amplitude is 0.11 multiplied by 0.43, which equals 0.047. After amplitude attenuation, the color gamut coverage compensation instruction is updated to compensate for an increase of 0.20% in color gamut coverage, and the brightness uniformity compensation instruction is updated to compensate for a decrease of 0.047 in brightness uniformity. The stress adjustment instruction maintains its original adjustment amplitude. The final collaborative control strategy includes four instructions: an increase of 1.41 micrometers in the stress peak position coordinate adjustment instruction, an increase of 0.34 MPa in the stress distribution uniformity standard deviation adjustment instruction, an increase of 0.20% in the color gamut coverage compensation instruction, and a decrease of 0.047 in the brightness uniformity compensation instruction.

[0075] This invention achieves automatic identification of conflicting command pairs by constructing an command influence matrix and prioritizes them based on path cumulative weights, thus resolving the antagonistic influence problem between stress control and optical compensation. The dynamic attenuation mechanism determines the attenuation coefficient based on the path weight ratio, ensuring the effective implementation of high-priority strategies while minimizing the negative impact of low-priority adjustments, thereby optimizing the overall control performance.

[0076] A second aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0077] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0078] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for synergistic control of surface stress distribution and optical performance of high color gamut privacy film, characterized in that, include: Obtain multilayer structural data of privacy films, and construct stress state feature sets and optical performance feature sets; The stress state feature set and the optical performance feature set are jointly encoded by a deep neural network to obtain a stress-optics co-representation vector. This co-representation vector embeds stress evolution trend information and optical response sensitivity information in a unified feature space. An attention mechanism is used to identify the interaction dependency between stress state features and optical performance features, and a bidirectional influence weight matrix is ​​established. This bidirectional influence weight matrix represents the positive influence path weight of stress evolution on optical performance and the negative constraint path weight of optical optimization requirements on stress distribution. The collaborative representation vector is input into the policy generation network, which analyzes the positive influence path of stress modulation on optical performance and the negative constraint path of optical optimization on stress distribution based on the bidirectional influence weight matrix, and generates a collaborative control policy including stress modulation instructions and optical compensation instructions. A process parameter adjustment scheme is generated based on the aforementioned collaborative control strategy.

2. The method according to claim 1, characterized in that, The steps for obtaining multilayer structural data of privacy films and constructing stress state feature sets and optical performance feature sets include: The material type and thickness information of each functional layer are converted into physical property parameters and optical parameters, respectively. Based on the physical property parameters, the interlayer stress gradient and stress distribution uniformity are calculated to form a stress state feature set. Based on the optical parameters, the light intensity distribution and color gamut coverage under different viewing angles are calculated, and the privacy viewing angle response features and color gamut boundary features are extracted to form an optical performance feature set.

3. The method according to claim 1, characterized in that, The steps of jointly encoding the stress state feature set and the optical performance feature set using a deep neural network to obtain the stress-optics co-representation vector include: A collaborative state mapping space with stress distribution state and optical performance state as two dimensions is constructed, and the state trajectory points of historical iterations are recorded; the collaborative optimal region is determined based on the set of points in the state trajectory points where stress deviation and optical performance deviation simultaneously satisfy the threshold condition. The stress state feature set is subjected to time-series unfolding to extract stress evolution trend information; the optical performance feature set is subjected to perturbation testing to extract optical response sensitivity information. The stress state feature set and the optical performance feature set are jointly encoded by the encoding layer of a deep neural network, and the stress evolution trend information and optical response sensitivity information are embedded to obtain an initial collaborative characterization vector. The initial collaborative representation vector is mapped to the collaborative state mapping space to obtain the current state point, and the deviation between the current state point and the collaborative optimal region is calculated. When the deviation exceeds the collaborative tolerance threshold, the initial collaborative representation vector is decomposed into stress-independent components and optical-independent components. After applying reverse compensation based on the deviation contribution of each independent component, the components are re-fused to obtain the collaborative representation vector.

4. The method according to claim 3, characterized in that, The steps of calculating the deviation between the current state point and the optimal cooperative region, and decomposing the initial cooperative representation vector into stress-independent components and optical-independent components when the deviation exceeds the cooperative tolerance threshold, applying reverse compensation based on the deviation contribution of each independent component, and then re-fusing to obtain the cooperative representation vector include: The Euclidean distance from the current state point to the ideal cooperative state reference point of the cooperative optimal region is calculated as the initial deviation. When the current state point is located outside the boundary of the cooperative optimal region, the shortest penetration distance to the boundary is calculated, and the weighted sum of the initial deviation and the penetration distance is used as the final deviation. When the final deviation exceeds the cooperative tolerance threshold, the initial cooperative characterization vector is decomposed into statistically independent stress independent components and optical independent components using the independent component analysis algorithm; the final deviation is decomposed into stress dimension deviation components and optical dimension deviation components, and the contribution ratio of each independent component to the corresponding dimension deviation component is calculated. A compensation vector pointing to the projection of the ideal cooperative state reference point onto the stress subspace is applied to the stress-independent component, and a compensation vector pointing to the projection of the ideal cooperative state reference point onto the optical subspace is applied to the optical-independent component. The compensation amplitude is modulated by the contribution ratio of each independent component. The compensated stress-independent component and optical-independent component are fused to obtain the cooperative characterization vector.

5. The method according to claim 1, characterized in that, The steps for identifying the interaction dependency between stress state characteristics and optical performance characteristics through an attention mechanism and establishing a two-way influence weight matrix include: The stress state feature set is converted into a query vector, and the optical performance feature set is converted into a key vector and a value vector. The correlation strength is calculated by the inner product operation of the query vector and the key vector. The correlation strength is used as a weight to perform weighted aggregation on the value vector to obtain a feature representation containing interactive dependencies, and an interactive dependency matrix is ​​constructed. Identify the positive influence path of stress state characteristics on optical performance characteristics from the interaction dependency matrix. The positive influence path represents the transmission relationship from stress evolution to optical performance change. Calculate the transmission weight coefficient of each feature node in the positive influence path. Identify the reverse constraint path of optical performance characteristics on stress state characteristics from the interaction dependency matrix. The reverse constraint path represents the feedback relationship between optical optimization requirements and stress distribution adjustment. Calculate the feedback weight coefficient of each feature node in the reverse constraint path. A two-way influence weight matrix is ​​constructed based on the transmitted weight coefficient and the feedback weight coefficient.

6. The method according to claim 1, characterized in that, The steps of inputting the collaborative representation vector into the policy generation network, which analyzes the positive influence path of stress modulation on optical performance and the negative constraint path of optical optimization on stress distribution based on the bidirectional influence weight matrix, and generating a collaborative control policy including stress modulation instructions and optical compensation instructions, include: The collaborative representation vector is input into the policy generation network. Based on the stress evolution trend information and optical response sensitivity information embedded in the collaborative representation vector, the stress distribution deviation vector and optical performance deviation vector between the current state and the target state are analyzed. Based on the transfer weight coefficients in the bidirectional influence weight matrix, the influence contribution of the stress distribution deviation vector on each optical performance feature is traced to construct a path diagram of the effect of stress regulation on optical performance; based on the feedback weight coefficients in the bidirectional influence weight matrix, the constraint strength of the optical performance deviation vector on each stress state feature is traced to construct a path diagram of the reverse constraint of optical optimization on stress distribution. Stress control nodes whose influence contribution exceeds the contribution threshold are extracted from the action path diagram, and stress control instructions are generated according to the weight distribution of the stress control nodes; optical compensation nodes whose constraint strength exceeds the constraint threshold are extracted from the reverse constraint path diagram, and optical compensation instructions are generated according to the weight distribution of the optical compensation nodes. Conflict detection is performed on stress control commands and optical compensation commands. When antagonistic influences are detected, priority ranking and coordination correction are performed based on the path weights of the action path diagram and the reverse constraint path diagram to obtain a cooperative control strategy.

7. The method according to claim 6, characterized in that, The steps for conflict detection between stress control commands and optical compensation commands, and prioritization and coordination correction based on path weights in the action path diagram and reverse constraint path diagram when antagonistic influences are detected, include: Construct an influence matrix for stress control commands and optical compensation commands, where each element represents the mutual influence coefficient between corresponding command pairs; identify command pairs with negative mutual influence coefficients from the influence matrix and mark them as conflicting command pairs; For each conflicting command pair, extract the path containing the stress control command in the conflicting command pair from the action path diagram, and calculate the cumulative path weight of the path; extract the path containing the optical compensation command in the conflicting command pair from the reverse constraint path diagram, and calculate the cumulative path weight of the path; compare the cumulative path weights of the two paths, and set the command with the larger cumulative path weight as the high priority command. The lower priority instruction in the conflicting instruction pair is attenuated by a factor that is determined by the ratio of the cumulative path weight of the higher priority instruction to the cumulative path weight of the lower priority instruction.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images