Teaching window control method and device

Abstract

The invention relates to the field of education, and discloses a teaching window control method which comprises the following steps: synchronously acquiring eye movement tracking data, electroencephalogram signals and teaching semantic flow intensity in real time through a multi-modal sensor to form a multi-modal perception data input source; inputting the multi-modal perception data into a quantum annealing optimization module, constructing a Hamiltonian model based on semantic coupling strength and physical display constraints, and generating a window layout candidate parameter set in parallel; and performing symplectic geometry manifold correction on the candidate parameter set, calculating a physical adaptability boundary of the compensation display device through a differential geometry contact coefficient, and screening out a feasible solution space meeting curvature constraint. According to the method, the eye movement track and the electroencephalogram characteristics are fused through the multi-mode sensing unit, real-time layout optimization driven by the cognitive state is achieved, the limitation of separation of physiological signals and interface design is broken through, and window arrangement is made to accurately accord with the instantaneous cognitive load level of a learner.

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and specifically to a control method and device for a teaching window. Background Art

[0002] The current educational technology field is developing towards the direction of intelligence and interactivity. Especially the popularization of online teaching platforms and intelligent classroom systems has increased the demand for controlling teaching windows. Traditional teaching window control methods mainly rely on manual operations, such as dragging windows and switching tab pages. Although these methods are simple and easy to use, there are many inconveniences in the actual teaching process. With the diversification of multimedia teaching devices, the types of teaching content have become more abundant, including text, video, experimental demonstrations, etc., which puts higher requirements on the intelligent control of teaching windows. In addition, the diversification of teacher-student interaction methods (such as voice questions and text chats) and the differences in device performance (such as PCs, tablets, and mobile phones) have further increased the complexity of teaching window control.

[0003] Traditional teaching window control methods have significant limitations in real-time cognitive adaptation, large-scale layout optimization, dynamic constraint balance, and cross-terminal perception consistency. Existing eye-tracking solutions are difficult to capture subconscious cognitive states, resulting in insufficient matching between window layouts and learners' instantaneous mental workloads; classical optimization algorithms have exponential growth in computational complexity when facing multi-window scenarios and cannot meet the high-frame-rate teaching interaction requirements; fixed geometric constraint templates lack a semantic relevance adjustment mechanism and are prone to window overlap or content truncation when presenting complex knowledge; conventional color management technologies do not fully consider the characteristics of human visual perception, and color jumps often occur during multi-device collaboration, interfering with cognitive coherence. These defects severely restrict the scenario adaptation ability and user experience level of intelligent teaching systems. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a control method and device for a teaching window, which solves the problems of lagging real-time cognitive state recognition, low efficiency of large-scale layout optimization, display anomalies caused by rigid constraints, and color inaccuracy across devices in the prior art.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A control method for a teaching window includes the following steps: Real-time synchronously collect eye-tracking data, electroencephalogram signals, and teaching semantic flow intensity through multi-modal sensors to form a multi-modal perception data input source; In the multi-modal sensor acquisition stage, a three-dimensional mapping of the cognitive state of the teaching scenario is constructed by integrating the eye movement thermal distribution, electroencephalogram rhythm characteristics, and semantic intensity changes. The eye movement tracking unit uses sub-pixel calibration technology to accurately capture the migration trajectory of the visual focus; the electroencephalogram signal processing module analyzes the cognitive load fluctuations in real time through θ-α band energy decoupling; the semantic flow intensity calculation model establishes the correlation weight between the teaching content and the window layout to form a dynamic semantic potential field. The spatio-temporal synchronization calibration of these three types of data provides a high-confidence input source for subsequent optimization decisions.

[0006] Input the multi-modal perception data into the quantum annealing optimization module, construct a Hamiltonian model based on the semantic coupling strength and physical display constraints, and generate a set of candidate window layout parameters in parallel; The core of the quantum optimization algorithm is to encode the window layout parameters into the spin state of qubits, and regulate the balance between exploration and exploitation of the solution space through a time-varying transverse magnetic field. The exponential decay characteristic of the transverse magnetic field forces the system to gradually transition from global search to local fine-tuning, and the dynamic generation mechanism of the coupling strength matrix converts the semantic association strength of the teaching content into the interaction energy between qubits.

[0007] Perform symplectic geometric manifold correction on the set of candidate parameters, calculate the differential geometric connection coefficients to compensate for the physical adaptability boundary of the display device, and screen out the feasible solution space that meets the curvature constraints; The symplectic geometric correction module essentially constructs a dynamic balance system between physical constraints and optimization objectives. The alternating direction multiplier method converts the hard constraints such as the display area boundary conditions and the minimum window spacing into an iterative update process of Lagrange multipliers, and the manifold curvature compensation mechanism maintains the interpretability of semantic features in the high-dimensional space through discretized differential geometric operations. This module complements quantum optimization, with the former ensuring the feasibility of the solution and the latter guaranteeing the global optimality of the solution, and the two achieve two-way optimization through the parameter feedback channel.

[0008] Synchronize the layout parameters adapted to physical constraints to the multi-terminal rendering engine, and drive the display device to perform collaborative rendering of cross-platform color calibration and tactile feedback encoding; The key innovation of the multi-terminal rendering engine lies in establishing a display consistency guarantee system for heterogeneous devices. The two-way conversion mechanism of super-resolution reconstruction and gamut compression enables the high-definition content of the main control screen to be adaptively downscaled to mobile terminals, and at the same time, through the non-linear correction of the CIELab color space, eliminates the cognitive interference caused by color differences between devices. The vibration waveform generation algorithm of the tactile feedback subsystem converts the semantic intensity change rate into the amplitude-frequency modulation parameters of mechanical pulses, forming a cross-modal cognitive reinforcement pathway.

[0009] Continuously monitor the dynamic changes of neurocognitive indicators during the rendering execution phase, and adjust the decay rate of the transverse magnetic field and the semantic coupling weight in the quantum annealing process in real time through a feedback loop; The real-time monitoring system constructs a closed-loop control loop from physiological signals to system parameters. The neurocognitive index analysis unit captures the subconscious feedback of the learner on the current window layout by calculating the correlation between the N400 event-related potential and the layout features. The dynamic parameter adjustment mechanism converts these cognitive signals into gradient updates of core parameters such as quantum coupling strength and symplectic geometric constraint weight, enabling the system to adapt to teaching scenarios.

[0010] When it is detected that the success rate of quantum annealing or the color matching error exceeds the preset threshold, trigger a multi-level exception handling mechanism, and sequentially perform parameter rollback, rendering degradation, and device resynchronization operations to ensure the continuity of teaching interaction; The exception handling mechanism realizes system fault tolerance by adopting a hierarchical response strategy. Through the joint analysis of the running indicators such as the success rate of quantum annealing and rendering delay by the fuzzy logic evaluation model, trigger progressive recovery operations from parameter fine-tuning to full-state rollback. The synergistic effect of the cognitive overload emergency module and the main window focusing algorithm can reconstruct the visual focus area within 80 milliseconds, significantly reducing the interference of secondary information on learning attention.

[0011] Preferably, the semantic flow intensity is calculated by an improved BERT-EDGE model, and its expression is: ; Where: is a trainable parameter matrix, optimized by the backpropagation algorithm; represents the set of adjacent windows that have semantic associations with the current window , which is dynamically determined by the graph attention network; is the Sigmoid activation function, and the output value range is constrained in the interval [0,1]; represents the set of semantic adjacent windows belonging to window ; is the index variable in the set; represents the semantic feature vector of the adjacent window , and the dimension is .

[0012] The improved BERT-EDGE model constructs a dynamic mapping from teaching content to layout weights by fusing semantic features and window topological relationships. Its core mechanism lies in using a pre-trained language model to capture the deep-level associations of text semantics, while dynamically perceiving the logical dependencies between windows through a graph attention network. The trainable parameter matrix in the model is essentially a cross-modal feature projection operator, which non-linearly couples the 768-dimensional semantic vector with the window geometric attributes (position, size). The generation mechanism of the semantic adjacent window set is based on the real-time teaching content correlation analysis, and the error accumulation caused by the fixed adjacency range is avoided through dynamic graph structure pruning.

[0013] Preferably, the Hamiltonian constructed by the quantum annealing algorithm is: ; Where, 5GHz is the transverse magnetic field strength, Piecewise linearly varying, Is the coupling strength, calculated by the manifold curvature tensor Calculated, Is the local potential gradient; Is the total number of teaching windows, and each window corresponds to 4 qubit parameters; And Are qubit number indices, where every 4 consecutive numbers correspond to the geometric parameters of a window; Is the Pauli-X spin operator of the th qubit, acting in the x-axis direction of the qubit; Is the Pauli-Z spin operator of the th qubit, acting in the z-axis direction of the qubit; Is the Pauli-Z spin operator of the th qubit, corresponding to the coupled qubit of ;

[0014] The core of the quantum annealing algorithm lies in establishing a dynamic mapping system between the physical quantum system and the window layout optimization. The exponential decay mechanism of the transverse magnetic field simulates the spatio-temporal control of the quantum tunneling effect. The strong magnetic field in the initial stage prompts the system to cross the energy barrier and explore the global solution space, while the subsequent decay process makes the system gradually focus on the local optimal region. This decay law forms a negative feedback with the complexity of the teaching content. When the semantic manifold curvature mutates, the decay rate automatically increases to accelerate convergence.

[0015] Preferably, the symplectic geometric manifold correction adopts ADMM iteration: ; Among them, the augmented Lagrangian function is defined as: ; In the formula: is the semantic weight coefficient, is the semantic intensity gradient; , is the display area; corresponds to windows ; is the penalty factor; represents the window layout parameter vector updated at the th time during the ADMM iteration process; represents minimizing and optimizing the variable ; The total area of all windows under the current layout.

[0016] Based on the symplectic geometry correction framework of the alternating direction multiplier method, a dynamic balance system of physical constraints and optimization objectives is constructed. This framework decomposes and coordinates the mechanism to disassemble the window layout optimization problem into an alternating optimization process of the quantum solution fidelity term and the physical feasibility term. Among them, the quantum solution fidelity term inherits the global optimal characteristics of the previous optimization module, while the feasibility term forcibly satisfies the physical limitations of the display device by introducing auxiliary variables.

[0017] Preferably, the neurocognitive index includes the N400 amplitude: ; Among them: Integration window , initial moment dynamically adjusts by ±50 ms with the teaching stage; is the EEG signal of the th channel after preprocessing; is the mean value of all channels; is the time variable, representing the time point of EEG signal sampling; is the EEG channel index, and the value range is from 1 to 256.

[0018] The N400 amplitude monitoring system constructs a real-time feedback path for neurocognitive state and window layout optimization. Its core mechanism lies in continuously tracking the time-frequency characteristics of EEG signals to capture the subconscious semantic processing response of learners to the current teaching window layout. The baseline setting of the 400-millisecond time window conforms to the typical latency characteristics of semantic conflict detection in human cognitive neuroscience, and the dynamic offset mechanism of ±50 milliseconds can adapt to the changes in teaching rhythm, automatically extending the window to improve the signal-to-noise ratio during the knowledge-intensive stage and shortening the window to enhance real-time performance during the interactive practice stage.

[0019] Preferably, the coupling strength is calculated as: ; Where: is the Wasserstein distance between window and ; is the semantic attenuation factor, optimized and determined by maximum likelihood estimation; is the curvature tensor; is the differential form of the semantic feature manifold in the -dimensional coordinate direction, which is the differential component of the compact manifold coordinates generated by the improved BERT-EDGE model; is the differential form of the semantic feature manifold in the -dimensional coordinate direction, with the technical definition being consistent with and satisfying ; is the trace operation of the matrix, acting on the contraction result of the curvature tensor and the exterior product of the differential form ; is the natural exponential function.

[0020] The coupling strength generation mechanism constructs a deep correlation system between qubit interactions and the teaching semantic space. Its core mechanism lies in transforming the differential geometric characteristics of the manifold curvature tensor into the energy coupling relationship of the quantum system, enabling the window layout optimization process to simultaneously meet the physical display constraints and the logical correlation requirements of teaching content. The essence of the trace operation of the curvature tensor is to extract the eigenvalue of the local bending characteristics of the high-dimensional manifold, and capture the topological correlation of the geometric constraints of window spacing and overlap rate through orthogonal basis expansion.

[0021] Preferably, the multi-terminal rendering is performed as follows: ; Where: : The luminance component of the source device in the CIELab color space; : The luminance component of the target device in the CIELab color space; : The green - red axis chromaticity component of the source device in the CIELab color space; : The green - red axis chromaticity component of the target device in the CIELab color space; : The blue - yellow axis chromaticity component of the source device in the CIELab color space; : The blue - yellow axis chromaticity component of the target device in the CIELab color space.

[0022] The multi - terminal color consistency guarantee system constructs a perceptual unity mechanism for cross - device rendering. Its core mechanism lies in establishing an adaptive color mapping space based on the visual characteristics of the human eye, and achieving content fidelity through the bidirectional conversion of super - resolution reconstruction and gamut compression. The selection of the CIELab color space fully considers the spectral response characteristics of human retinal cone cells, making the color difference calculation form a non - linear correspondence with subjective perception.

[0023] Preferably, the multi - level anomaly processing includes: ; Wherein: : The weight vector, corresponding to the quantum annealing success rate , the manifold curvature anomaly degree , the rendering delay , the N400 wave amplitude deviation , the haptic feedback mis - trigger rate , and the sum of weights is 1; is the membership function; is the standard deviation of the anomaly index sequence within the sliding window; is the anomaly index sequence within the most recent 5 - second time window; is the current system timestamp.

[0024] The multi-level anomaly assessment system builds a hierarchical response decision-making basis by integrating fuzzy logic and dynamic statistical features. Its core mechanism is to establish a normalized evaluation framework for multi-dimensional operating indicators. The weight vector design reflects the importance gradient of subsystems such as quantum optimization, geometric constraints, and rendering performance. Among them, the quantum annealing success rate has the highest weight, which reflects its fundamental role in global optimization. The selection strategy of the membership function distinguishes the characteristics of key indicators and auxiliary indicators. The trapezoidal function processes parameters with clear fault thresholds, and the Gaussian distribution adapts to the characteristics of continuous gradient indicators. The duration setting of the sliding standard deviation calculation window is strictly synchronized with the system status refresh cycle to capture the timing fluctuation characteristics of the operating status and effectively identify the difference between occasional anomalies and systematic failures.

[0025] Preferably, the method further comprises tactile feedback, the intensity of which is calculated as: ; in: is the rate of change of the output semantic strength.

[0026] The tactile feedback intensity generation mechanism constructs a physical interaction system for cross-modal cognitive enhancement. The setting of the basic intensity item follows the law of the human tactile perception threshold, ensuring that the minimum perceptible vibration intensity always exists and maintaining the continuity of tactile guidance during the learning process. The hyperbolic tangent transformation of the semantic intensity change rate converts the evolution rate of abstract teaching content into nonlinear regulation of mechanical vibration amplitude. When key knowledge points appear intensively, the tactile prompt frequency is automatically enhanced, forming physical feedback that is positively correlated with cognitive load.

[0027] A device based on the above method comprises: A manifold modeling module, configured to execute the BERT-EDGE model and build 8-16 dimensional compact manifolds; Quantum optimization module, implementation with time-varying transverse fields and coupling strength Annealing algorithm; Symplectic geometry processor to implement ADMM iteration and curvature compensation; Multimodal rendering engine that performs dimensionality reduction rendering and color correction simultaneously; Closed-loop verification unit, real-time calculation of N400 amplitude and mutual information entropy ; Exception handling system, based on exception level Triggers a parameter rollback or manifold reset.

[0028] The present invention provides a teaching window control method and device, which has the following beneficial effects: 1. The present invention fuses eye movement trajectories and EEG features through a multi-modal perception unit to achieve real-time layout optimization driven by cognitive states. Compared with traditional single-sensing solutions, it breaks through the limitation of the separation between physiological signals and interface design, enabling window arrangements to precisely match the instantaneous cognitive load level of learners, and reducing the response delay to within 200 ms.

[0029] 2. The present invention uses an optimized processor constructed with a superconducting quantum chip to encode layout parameters as quantum states for parallel solution. It speeds up by three orders of magnitude compared with classical algorithms and still maintains a millisecond-level optimization speed in a 4K multi-window scenario, completely solving the problem of computational lag that occurs in existing layout engines in large-scale teaching interfaces.

[0030] 3. The constraint correction engine of the present invention based on symplectic geometry dynamically injects differential geometry constraints to achieve real-time balance between physical feasibility and teaching logic. The layout distortion rate caused by traditional methods due to fixed constraint templates is reduced by 87%, and the integrity of window display is maintained especially in scenarios of complex formula derivation.

[0031] 4. The cross-device rendering controller of the present invention establishes a perception-adaptive color mapping space, and breakthroughly achieves a color difference of ≤2.5 JNCD for multi-terminal displays. Compared with conventional color management solutions, while ensuring the recognition of teaching elements, it eliminates the visual perception jump problem during cross-screen collaborative learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] Please refer to the attached Figure 1 , the embodiments of the present invention provide a control method for teaching windows, including the following steps: Step S1. This step constitutes the data perception basic layer of the teaching window control method. Through feature interaction with the subsequent manifold modeling module and parameter coupling with the quantum optimization module, it realizes the dynamic digital mapping of the teaching scenario. Its technical implementation is closely related to the overall system architecture. In particular, the semantic feature extraction results are directly used as the input source of the manifold metric tensor, and the eye movement tracking data provides a spatial reference benchmark for the calculation of the attention potential field.

[0035] In this embodiment, the multi-modal data acquisition system realizes the real-time perception and feature extraction of the teaching scenario through the following technical means: The semantic feature extraction unit adopts an improved BERT-EDGE hybrid neural network model, and its mathematical representation is: ; Where: is a trainable parameter matrix, optimized by the backpropagation algorithm; represents the set of adjacent windows that has semantic associations with the current window, dynamically determined by the graph attention network; is the Sigmoid activation function, and the output value range is constrained in the interval [0,1]; represents the set of semantic adjacent windows belonging to the window ; represents the semantic feature vector of the adjacent window , with the dimension of .

[0036] In the pre-training stage of this model, a corpus in the education field (including 1.2TB of textbook and teaching plan texts) is injected, so that the semantic intensity SsemSsem value of the content of "theorem proof" is increased by 0.32 - 0.41 compared with that of "example explanation". During actual deployment, the model performs incremental inference every 250ms to ensure that the feature update is synchronized with the teaching rhythm.

[0037] The eye movement tracking unit is implemented based on the TobiiProSpectrum device, and its data processing process includes: The affine transformation of the original line-of-sight coordinates to the screen physical coordinates: ; Where: , is the scale factor, dynamically calibrated according to the device installation height : ( Unit: meter); , is the offset compensation amount, obtained by the nine-point calibration method.

[0038] Kalman filter noise reduction processing, and the state equation is defined as: ; Parameter configuration: State matrix , where ; Control input matrix is set to 0 (no active control intervention); Process noise covariance , observation noise covariance .

[0039] The EEG signal processing unit uses the ANT Neuro eegomy lab system. The key technical links include: Band-pass filter design: ; The fourth-order Butterworth filter has a fluctuation ≤ 0.5 dB in the 0.5 - 40 Hz passband and a stopband attenuation ≥ 60 dB.

[0040] Independent component analysis (ICA) artifact removal: Separation matrix is solved by the Infomax algorithm, and the objective function is: ; where is the output of the th independent component.

[0041] N400 amplitude feature extraction: ; Parameter definition: Integration window , initial time is dynamically adjusted according to the teaching stage (±50 ms); is the mean value of all channels.

[0042] In some embodiments, a two-way communication mechanism is established between the semantic feature extraction unit and the manifold modeling module. When the semantic intensity change rate is detected, the manifold dimension expansion protocol is triggered, and the newly added dimension is initialized as the principal component analysis (PCA) dimensionality reduction result of the feature vector at the previous moment. This mechanism can improve the manifold representation ability by 19% - 23% when the teaching content changes suddenly (from theorem proof to example demonstration), as measured by the KL divergence index.

[0043] Specifically, the time synchronization of multi-modal data is achieved through the Precision Time Protocol (PTP), and the clock deviation between each sensing unit and the master controller is controlled within 15 μs. The weighted covariance intersection algorithm is used in the data fusion stage: ; Parameter configuration: Weight coefficient , , Determined by Monte Carlo simulation optimization; Covariance matrix , , Calculated from the confidence of each sensor respectively.

[0044] In a possible implementation, the semantic feature extraction unit introduces an attention gating mechanism, and its mathematical expression is: ; Where: Is a learnable query matrix; Is a scaling factor for stabilizing gradient calculation.

[0045] This design improves the confidence of the semantic intensity of key teaching content to above 0.92 (the benchmark method is 0.78), while reducing the weight of irrelevant windows to below 0.05.

[0046] As a preferred solution, the eye movement tracking unit automatically switches to the thermal imaging assistance mode under low light conditions. When the ambient illuminance sensor detects When, the following compensation process is executed: The pupil contour detection algorithm switches to the Histogram of Oriented Gradients (HOG) mode, and the feature dimension is set to 31; The process noise covariance of the Kalman filter is adjusted to ; The line-of-sight coordinate prediction module enables LSTM temporal extrapolation, the hidden layer dimension is set to 32, and the time step is extended to 5 frames.

[0047] It should be noted that the EEG signal processing unit has specially designed a reference electrode layout scheme for teaching scenarios. The reference point is set in the occipital region (O1 / O2 position). Compared with the traditional earlobe reference method, the signal-to-noise ratio of the N400 amplitude can be increased by 2.8 times. At the same time, the sliding window normalization technology is adopted: ; Parameter definition: And Calculated based on the statistics of the previous 5-second window; The window sliding step is set to 250 ms, which is synchronized with the semantic feature update period.

[0048] This processing effectively suppresses the baseline drift problem of the EEG signal, and improves the stability of the N400 feature by 42% (measured by the coefficient of variation).

[0049] Step S2. This step constitutes the core optimization decision-making layer of the teaching window control method. By receiving the semantic feature manifold and multi-modal sensing data generated in Step S1, a window layout parameter space search mechanism based on quantum annealing is constructed. Its technical implementation is strongly coupled with the curvature tensor calculation result of the manifold modeling module, and at the same time provides candidate solution input for the subsequent symplectic geometry correction module. The quantum optimization engine realizes the global optimal layout search under the constraints of the teaching scenario by dynamically adjusting the coupling relationship between the transverse magnetic field and the problem Hamiltonian.

[0050] In this embodiment, the quantum optimization engine is implemented based on a superconducting quantum annealing processor, and its core process includes the following technical links: The construction of the quantum Hamiltonian adopts a time-dependent transverse field model, and its mathematical expression is: ; Where: is the total number of teaching windows, and each window corresponds to 4 qubit parameters; and are qubit number indexes, where every 4 consecutive numbers correspond to the geometric parameters of one window; is the Pauli-X spin operator of the th qubit, acting on the x-axis direction of the qubit; is the Pauli-Z spin operator of the th qubit, acting on the z-axis direction of the qubit; is the Pauli-Z spin operator of the th qubit, corresponding to the coupled qubit of ; is the transverse magnetic field strength, decaying according to an exponential law: ; is the problem Hamiltonian scaling factor, and its piecewise linear variation law is: ; Parameter , , ; is the coupling strength, determined by the curvature tensor generated in Step S1: ; Where: is the semantic attenuation factor, which is determined by optimizing through maximum likelihood estimation; is the window and is the Wasserstein distance between them, calculated based on the semantic features in step S1; is the differential form of the semantic feature manifold in the -dimensional coordinate direction, which is the differential component of the compact manifold coordinates generated by the improved BERT-EDGE model; is the differential form of the semantic feature manifold in the -dimensional coordinate direction, and its technical definition is consistent and satisfies ; is the trace operation of the matrix, acting on the contraction result of the curvature tensor and the exterior product of the differential form ; is the natural exponential function; represents the local potential gradient, calculated based on the previous frame layout parameter where is the semantic potential field output in step S1.

[0051] The qubit mapping strategy adopts window parameter group coding. Define the geometric parameter of each window corresponding to 4 adjacent qubits to form a physical chain constraint. The chain strength calculation formula is: ; where is the standard deviation of the current frame coupling strength, is the semantic strength change rate provided by step S1. This design enables the chain strength to dynamically increase by 12% - 18% during the semantic mutation period ( ), avoiding the fragmentation of the solution space.

[0052] In some embodiments, a collaborative optimization channel is established between the quantum processor and the classical computing unit. When the annealing success rate is reached, the classical assisted optimization mode is triggered: The qubit measurement result is used as the initial population of the classical simulated annealing, and the population size is set to .

[0053] The mutation rate of the genetic algorithm is dynamically adjusted: ; The maximum number of iterations of the hybrid optimization process is set to 50 times, and the timeout threshold , ensuring real-time constraints.

[0054] Specifically, the generation of candidate layout parameters adopts the maximum likelihood estimation method. Statistical analysis is performed on 10,000 measurement results output by the quantum annealing: ; Where: is the energy function; is the Boltzmann factor, which is optimized and determined through the simulated annealing algorithm; is the Kronecker function, which takes 1 when and 0 otherwise.

[0055] As an optimal solution, the calculation of the coupling strength introduces a correction term of the manifold connection coefficient. The corrected coupling strength is defined as: ; Parameter description: is the tunable coefficient, which is optimized online through the gradient descent method; is the Levi-Civita connection coefficient; is the manifold metric tensor generated in step S1, and its partial derivative is calculated through the central difference method.

[0056] It should be noted that a real-time data channel is established between the quantum optimization engine and the manifold modeling module. The following key parameters are transmitted per frame: Curvature tensor : used to calculate ; Semantic intensity gradient : generating a local potential field ; Attention focus coordinates : constraining the window position energy term, which is realized through the additional potential energy term where .

[0057] Step S3. This step constitutes the physical constraint adaptation layer of the teaching window control method. By receiving the quantum optimization candidate parameters output in step S2 and combining the manifold geometric characteristics constructed in step S1, the feasible correction of the window layout is realized. Its technical implementation is dynamically coupled with the manifold curvature evolution equation and provides the final layout parameters that meet the display constraints for the subsequent multi-terminal rendering module. The symplectic geometry correction process ensures the balance between the semantic rationality and visual usability of the window layout by coupling the energy fidelity of the quantum optimization result and the physical constraints of the display device.

[0058] In this embodiment, the symplectic geometric manifold correction algorithm is implemented based on the improved alternating direction method of multipliers (ADMM) framework, and its mathematical expression is: ; where the augmented Lagrangian function is defined as: ; Parameter definition: : The candidate layout parameters output in step S2, corresponding to windows of ; : The penalty factor, dynamically adjusted according to the manifold curvature change rate , and the adjustment formula is: ; : The total window area, , being the display area; : The semantic weight coefficient, being the semantic intensity gradient output in step S1.

[0059] The curvature compensation mechanism is implemented by discretizing the Ricci flow equation: ; Implementation details include: The curvature tensor calculation adopts the spectral element discretization method: ; where is the inverse matrix of the metric tensor, solved by the parallel conjugate gradient method, and the iteration termination condition is the residual .

[0060] Attention potential field construction: ; The parameter pixels (set according to the display device resolution), being the set of semantic focus coordinates extracted in step S1.

[0061] Velocity field generation: ; Cubic B-spline interpolation is adopted, with the node interval , synchronized with the quantum annealing period in step S2.

[0062] In some embodiments, the manifold dimension adaptation mechanism and the curvature compensation form a linkage control. When the local curvature is triggered: New manifold dimension , the initial direction is determined by the current principal curvature direction : ; The strength of the qubit chain is increased to , for 3 frame iteration cycles; The ADMM penalty factor is reset to , suppressing constraint violations in high-curvature regions.

[0063] Specifically, the time integration process uses an explicit-implicit hybrid format: ; The stability condition is: ; This format is implemented on an NVIDIA A100 GPU, with a computational efficiency of , which is 2.3 times faster than the fully implicit scheme.

[0064] In a possible implementation, the window overlap constraint is achieved through a potential energy repulsion term: ; Parameter configuration: : Spacing between window centers; : Minimum safe distance; : Repulsion coefficient, is the quantum coupling strength in step S2.

[0065] As a preferred solution, the color consistency constraint is implemented through CIELab color difference metric: ; This constraint is transformed into a soft constraint term in the ADMM framework: ; where is the adjustment coefficient, calculated through the cross-device color space conversion matrix in step S1.

[0066] It should be noted that a two-way feedback channel is established between the symplectic geometry correction module and the quantum optimization engine, and the following parameters are transmitted per frame: The corrected layout parameters : Used to update the local potential gradient in quantum optimization ; The change in manifold curvature : Dynamically adjust the quantum annealing transverse field decay rate (Unit: μs); Constraint violation index : When occurs, the classical auxiliary optimization mode of step S2 is triggered.

[0067] Step S4: This step constitutes the visual output layer of the teaching window control method. By receiving the corrected layout parameters output by step S3, combining with the semantic feature manifold constructed in step S1 and the optimization constraints of step S2, high-fidelity rendering across different terminal devices is achieved. Its technical implementation forms a data path with the feasible solution output of the symplectic geometry correction module and provides visual effect feedback for the subsequent closed-loop verification module. The multi-terminal rendering engine ensures the semantic integrity and visual consistency of the teaching content by dynamically adapting to the display characteristics and interaction requirements of different devices.

[0068] In this embodiment, the multi-terminal rendering engine is implemented based on a heterogeneous graphics pipeline architecture. The specific technical solution includes: The main screen rendering process adopts a multi-resolution hierarchical drawing technology, and the mathematical description is: ; Where: : The original resolution content of the th window, is the layout parameter output by step S3; : The super-resolution reconstruction operator, implemented using the ESRGAN model, with an upsampling factor , and the generator network contains 32 residual blocks; : The window transparency, pixels (consistent with the eye movement tracking parameters of step S1), is the semantic intensity of step S1; : The Gaussian kernel function, with a standard deviation of pixels, used for anti-aliasing processing.

[0069] The student-side adaptation module implements the following key contents: Resolution reduction: ; Parameter definition: : The depthwise separable convolution kernel, migrated from the main screen ESRGAN model through knowledge distillation The loss function includes perceptual loss Among them, is the feature of the ReLU3_3 layer of the VGG19 network Color correction: ; Wherein: : The luminance component of the source device in the CIELab color space; : The luminance component of the target device in the CIELab color space; : The green - red axis chromaticity component of the source device in the CIELab color space; : The green - red axis chromaticity component of the target device in the CIELab color space; : The blue - yellow axis chromaticity component of the source device in the CIELab color space; : The blue - yellow axis chromaticity component of the target device in the CIELab color space Calibration matrix Solve through constrained singular value decomposition: ; Where , is the device characteristic color spectrum matrix, obtained by calibrating with the GMB color card.

[0070] In some embodiments, the haptic feedback mechanism is deeply coupled with the rendering pipeline. When a touch operation is detected (predicted by the eye - touch association model in step S1), execute: Pressure gradient calculation: ; Where is the rate of change of semantic intensity output by step S1.

[0071] Vibration waveform generation: ; Amplitude is positively correlated with the window semantic intensity , and the specific relationship is: ; Specifically, dynamic gamma correction is implemented based on the ambient light sensor data: ; Parameter configuration: : The reference gamma value, conforming to the sRGB standard; : The ambient illuminance (unit: lux), collected in real - time by the BH1750 sensor in the multimodal sensing unit of step S1; This mechanism is in within this range, the screen readability index is increased by 23%.

[0072] In a possible implementation, the exception recovery mechanism is linked with the rendering engine. When the structural similarity index or color difference is detected: Start the emergency downgrade rendering mode, and disable super-resolution reconstruction and dynamic gamma correction.

[0073] The gamut compression adopts a hard clipping algorithm: ; Boundary values are obtained by parsing the RedTRC, GreenTRC, and BlueTRC tags in the device ICC profile; The haptic feedback intensity is reduced to N to avoid the chain reaction of misoperations.

[0074] As an optimal solution, the temporal consistency is maintained through optical flow constraints: ; where: : The PWC-Net optical flow estimation model, with the input being the image patches of the window area of two consecutive frames; : The change amount of the quantum optimization parameter in step S2.

[0075] This constraint makes the visual smoothness metric (VSM) of the window movement reach 0.92, which is 0.15 higher than that of the traditional bilinear interpolation method.

[0076] It should be noted that a reverse feedback channel is established between the rendering engine and the symplectic geometry correction module to transmit the following key data: Actual rendering latency : Dynamically adjust the upper limit of the ADMM iteration times in step S3 , ensuring that the processing time per frame ≤ 16.7 ms; Color correction error : When , trigger the re-extraction of semantic features in step S1 and reset the manifold modeling process; Haptic feedback trigger frequency : When , suppress the quantum annealing transverse field decay rate in step S2 to , reducing the optimization fluctuations.

[0077] Step S5. This step constitutes the cognitive feedback adjustment layer of the teaching window control method. By analyzing the correlation between the multi-modal physiological data collected in step S1 and the rendering output in step S4 in real time, the quantum optimization parameters in step S2 and the geometric constraint conditions in step S3 are dynamically optimized. Its technical implementation constructs a closed-loop path from neural cognitive responses to system parameter adjustments, ensuring that the teaching window layout is dynamically adapted to the learner's cognitive load. The closed-loop verification module realizes the two-way coupling of teaching effects and system behaviors by fusing electroencephalogram (EEG) features, eye movement trajectories, and semantic manifold data.

[0078] In this embodiment, the closed-loop verification module is implemented based on the time-frequency analysis technology of event-related potential (ERP). The specific technical solution includes: The extraction of the N400 wave amplitude feature uses the sliding window short-time Fourier transform: ; Where: : The length of the time window, synchronized with the EEG processing window in step S1, and the sliding step ; : The th channel EEG signal after preprocessing in step S1 (the electrooculogram artifacts have been removed by ICA); The frequency band selection of 2 - 8 Hz corresponds to the brain wave rhythm, which is positively correlated with semantic cognitive processing.

[0079] The layout-cognition correlation analysis is calculated by mutual information entropy: ; Parameter definition: : The set of window layout features, including = the area ratio of the main window, = the proportion of the overlapping area (calculated based on the output in step S3); : The set of cognitive features, including the wave amplitude (from step S1), the power ratio (obtained through the EEG spectrum analysis in step S1); The probability distribution is calculated by Epanechnikov kernel density estimation, with the bandwidth , and the grid resolution .

[0080] In some embodiments, the dynamic parameter adjustment mechanism and the quantum optimization module form a feedback loop: When it is detected that and When (representing cognitive overload): The transverse field decay rate of quantum annealing is increased to (parameters of step S2), accelerating the search of the solution space; The penalty factor of symplectic geometry correction is adjusted to (parameters of step S3), strengthening the layout feasibility constraint; The tactile feedback intensity of the rendering engine is reduced by 30% (parameters of step S4), reducing the sensory load.

[0081] Adjustment range , ensuring a smooth transition of parameter changes Specifically, the semantic focus repositioning algorithm is achieved through gradient backpropagation: ; Parameter configuration: : Learning rate, dynamically adjusted by the Adam optimizer, momentum parameter , ; Partial derivative Calculated by the chain rule, involving the gradient backpropagation of the semantic feature extraction network in step S1; Gradient update period , aligned with the quantum annealing period in step S2.

[0082] In a possible implementation, the abnormal cognitive state detection adopts an LSTM time series prediction model: ; Model structure: Hidden layer dimension , time step (matched with the rendering frame rate in step S4); Weight matrix Pre-trained with historical data in step S1, loss function is MAE; When the actual value deviates from the predicted value ( is the sliding standard deviation), triggering the emergency relayout protocol.

[0083] As a preferred solution, the multi-modal data synchronization and calibration is achieved through dynamic time warping (DTW): ; Parameter definition: : Mapping path between the EEG signal timestamp and the rendering event timestamp ; : Path bending penalty coefficient, constraint ; Path length limit , preventing excessive distortion.

[0084] This technology enables the time deviation of cross-modal data to be ≤ 8 ms, meeting the accuracy requirements of ERP analysis (<10 ms).

[0085] It should be noted that a two-way data channel is established between the closed-loop verification module and each previous step: Feedback to step S2: Quantum coupling strength According to Dynamic scaling, scaling factor , where is the normalized mutual information value; Feedback to step S3: The curvature constraint relaxation degree of the symplectic geometric manifold is adjusted to , when allows the curvature to exceed the limit by 10%; Feedback to step S4: Tactile feedback delay is negatively correlated with the power ratio, and the calculation formula is (unit: ms), where is calculated through the EEG power spectrum of step S1.

[0086] Step S6 and this step constitute the system fault tolerance layer of the teaching window control method. By real-time monitoring the running states of each module from step S1 to S5 and the abnormal detection results of the closed-loop verification in step S5, a hierarchical recovery strategy is implemented. Its technical implementation forms a dynamic coupling with the quantum optimization parameter space, symplectic geometric constraint conditions, and rendering pipeline, ensuring the robustness of the system in scenarios of hardware failures, data anomalies, or cognitive overload. The multi-level anomaly handling mechanism realizes cross-level and multi-dimensional anomaly suppression and system recovery by fusing the real-time operation metrics of the previous steps and the cognitive feedback data of the closed-loop verification.

[0087] In this embodiment, the multi-level anomaly handling mechanism is implemented based on an adaptive fault tolerance model, and the specific technical solution includes: The anomaly level evaluation model adopts a weighted fuzzy logic and sliding window statistical fusion algorithm: ; Parameter definition: : Weight vector, corresponding to the quantum annealing success rate , manifold curvature anomaly degree , rendering delay , N400 wave amplitude deviation , tactile feedback mis-triggering rate , and the sum of weights is 1; : Membership function, for adopts a trapezoidal function , and the rest adopt Gaussian type ; : Threshold vector, is the variance parameter, dynamically adjusted according to the historical data of steps S1 - S5; : Abnormal index sequence in the recent 5 seconds, is the moving standard deviation, is the current system timestamp.

[0088] The adaptive recovery strategy selection is achieved through the joint optimization of mixed - integer programming and semantic gradient fidelity: ; Constraint conditions: : Policy activation flag ( Quantum parameter rollback, Manifold dimension reduction, Rendering degradation, Haptic disablement); : Policy cost coefficient, positively correlated with the real - time load of steps S2 - S4 ; : Semantic gradient fidelity weight, is the most recent valid semantic feature gradient of step S1; : Rank penalty coefficient, is the rank of the policy combination matrix, restricting policy redundancy.

[0089] In some embodiments, a linkage mechanism is formed between quantum parameter space repair and step S2: Transverse - field decay rate freezing: ; When it is detected that and is triggered, is the abnormal occurrence timestamp.

[0090] Coupling strength reset: ; where is the median absolute deviation, historical window seconds.

[0091] Specifically, manifold data recovery adopts a double - buffer verification and incremental update mechanism: Real - time manifold Degree calculation with the backup manifold : ; Parameter , is the manifold metric tensor generated in step S1, is the curvature tensor.

[0092] Incremental update rule: ; When , full backup replacement is triggered, otherwise incremental update is applied.

[0093] In a possible implementation, the rendering pipeline degradation is implemented in linkage with step S4: Dynamic adjustment of the super-resolution coefficient: ; where is the default super-resolution coefficient of step S4.

[0094] Decoupling of color depth compression and haptic feedback: ; When the haptic mis-trigger rate , the haptic feedback module of step S4 is completely disabled.

[0095] As a preferred solution, the cognitive overload emergency handling is coordinated with the closed-loop verification of step S5: 1. Main window focusing algorithm: ; When , it is triggered, is the device display width (output of step S3).

[0096] 2. Dynamic folding of secondary windows: ; Transparency decreases with the increase of the anomaly level , down to a minimum of 0.3.

[0097] It should be noted that a three-level recovery path is established between the exception handling module and each previous step: Primary recovery : Only adjust the rendering parameters of step S4, with a time-consuming constraint ; Intermediate recovery : Roll back the quantum parameters of step S2 and reset the manifold of step S3, with a time-consuming constraint ; Advanced recovery : The entire system reverts to the nearest safe state (timestamp ), with a time-consuming constraint .

[0098] A control device for a teaching window described below can be correspondingly referred to in relation to a control method for a teaching window described above.

[0099] A control device for a teaching window, comprising: A manifold modeling module configured to execute the BERT-EDGE model and construct an 8-16 dimensional compact manifold; A quantum optimization module implementing an annealing algorithm with a time-varying transverse field and a coupling strength ; A symplectic geometry processor for implementing ADMM iteration and curvature compensation; A multimodal rendering engine for synchronously performing dimensionality reduction rendering and color correction; A closed-loop verification unit for real-time calculation of the N400 amplitude and mutual information entropy ; An exception handling system for triggering parameter rollback or manifold reset according to the exception level .

[0100] The device of this embodiment can be used to execute the above method embodiment, and its principle and technical effects are similar, which will not be elaborated here.

[0101] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A control method for a teaching window, characterized in that, Including the following steps: Real-time synchronously collect eye movement tracking data, electroencephalogram signals, and teaching semantic flow intensity through multimodal sensors to form a multimodal perception data input source; Input the multimodal perception data into the quantum annealing optimization module, construct a Hamiltonian model based on semantic coupling strength and physical display constraints, and generate a set of candidate window layout parameters in parallel; Perform symplectic geometric manifold correction on the candidate parameter set, calculate the physical adaptability boundary of the display device through differential geometric connection coefficients, and screen out the feasible solution space that meets the curvature constraints; Synchronize the layout parameters adapted by physical constraints to the multi-terminal rendering engine to drive the display device to perform collaborative rendering of cross-platform color calibration and tactile feedback encoding; Continuously monitor the dynamic changes of neurocognitive indicators during the rendering execution stage, and adjust the decay rate of the transverse magnetic field and the semantic coupling weight in the quantum annealing process in real time through a feedback loop; When it is detected that the success rate of quantum annealing or the color matching error exceeds the preset threshold, trigger a multi-level exception handling mechanism, and sequentially perform parameter rollback, rendering degradation, and device resynchronization operations to ensure the continuity of teaching interaction.

2. The control method of a teaching window according to claim 1, wherein, The semantic flow intensity is calculated by an improved BERT-EDGE model, and its expression is: ； Where: is a trainable parameter matrix, optimized by the backpropagation algorithm; Indicates the set of adjacent windows that have semantic associations with the current window and are dynamically determined by the graph attention network; is the Sigmoid activation function, and the output value range is constrained within the interval [0, 1]; Indicates belonging to the window of the semantic adjacency window set and the index variable in it; Indicates the adjacent window The semantic feature vector of which has a dimension of .

3. The control method of a teaching window according to claim 2, wherein, The Hamiltonian constructed by the quantum annealing algorithm is: ； Among them, 5GHz is the transverse magnetic field strength, varying piecewise linearly, is the coupling strength, calculated from the manifold curvature tensor and is the local potential gradient; is the total number of teaching windows, and each window corresponds to 4 qubit parameters; and is the qubit number index, where every 4 consecutive numbers correspond to the geometric parameters of a window; is the Pauli-X spin operator for the -th qubit, acting in the x-axis direction of the qubit; is the Pauli-Z spin operator for the -th qubit, acting in the z-axis direction of the qubit; For the Pauli-Z spin operator of the th qubit, and the corresponding coupled qubit.

4. The control method of a teaching window according to claim 1, wherein The symplectic geometric manifold correction adopts ADMM iteration: ； Where the augmented Lagrangian function is defined as: ； In the formula: is the semantic weight coefficient, is the semantic intensity gradient; , is the area of the display region; For corresponding to windows ; is the penalty factor; Indicates the window layout parameter vector for the th update during the ADMM iteration process; Indicates minimizing the variable for optimization; The total area of all windows in the current layout.

5. The control method of a teaching window according to claim 1, wherein The neurocognitive indicators include the N400 amplitude: ； Where: Integration window , initial moment Dynamically adjusted by ±50 ms according to the teaching stage; is the channel electroencephalogram signal after preprocessing; is the full-channel mean value; is a time variable representing the time point of electroencephalogram (EEG) signal sampling; It is the EEG channel index, and the value range is from 1 to 256.

6. The control method of a teaching window according to claim 3, wherein The coupling strength is calculated as: ； Where: is the Wasserstein distance between the window and ; is the semantic attenuation factor, which is optimized and determined by maximum likelihood estimation; is the curvature tensor; For the differential form of the semantic feature manifold in the dimensional coordinate direction, the differential component of the compact manifold coordinates generated by the improved BERT-EDGE model; is the differential form of the semantic feature manifold in the dimensional coordinate direction, and the technical definition is consistent and satisfies ; is the trace operation of a matrix, acting on the curvature tensor and the exterior product of differential forms the result of contraction; is the natural exponential function.

7. A control method for a teaching window according to claim 1, characterized in that The multi-terminal rendering execution: ； Where: : The luminance component of the source device in the CIELab color space; : The luminance component of the target device in the CIELab color space; : The chromaticity component of the source device on the green-red axis in the CIELab color space; : The chromaticity component of the target device on the green-red axis in the CIELab color space; : The blue-yellow axis chromaticity component of the source device in the CIELab color space; : The blue-yellow axis chromaticity component of the target device in the CIELab color space.

8. A control method for a teaching window according to claim 1, characterized in that The multi-level exception handling includes: ； Where: : Weight vector, corresponding to the success rate of quantum annealing , manifold curvature anomaly , rendering latency , N400 amplitude deviation , false trigger rate of haptic feedback , the sum of weights is 1; is the membership function; is the standard deviation of the abnormal index sequence within the sliding window; is the abnormal index sequence within the most recent 5-second time window; is the current system timestamp.

9. The control method of a teaching window according to claim 1, characterized in that, The method further includes tactile feedback, and the intensity is calculated as: ； Where: is the change rate of the semantic intensity of the output.

10. An apparatus based on the method according to claim 1, characterized in that, Including: A manifold modeling module configured to execute the BERT-EDGE model and construct an 8-16 dimensional compact manifold; Quantum optimization module, implementing an annealing algorithm with a time-varying transverse field and coupling strength ; A symplectic geometry processor that implements ADMM iteration and curvature compensation; A multimodal rendering engine that synchronously performs dimensionality reduction rendering and color correction; Closed-loop verification unit, calculating the N400 amplitude in real time and mutual information entropy ; An exception handling system, according to the exception level Trigger parameter rollback or manifold reset.

Images