Graphics workstation data transmission optimization method based on information security

Abstract

The invention discloses a graphic workstation data transmission optimization method based on information security. The method comprises the following steps: S1, acquiring to-be-transmitted graphic data and a transmission situation parameter set; s2, constructing a transmission situation feature field, inputting an enhanced Fourier neural operator model, and generating a path constraint matrix and a transmission control action feasible set; s3, carrying out joint coding processing; s4, jointly inputting an improved Decision Transform model, introducing a projection double-dual mechanism, and generating a cooperative transmission control action; s5, executing multi-path security scheduling transmission control; s6, calculating a path constraint margin field; and S7, updating the transmission situation characteristic field. According to the method, multi-path safe and controllable transmission of the data of the graphic workstation in a complex network environment is realized, and the feasibility, the adaptivity and the overall transmission efficiency of a transmission decision are improved while the transmission safety is ensured.

Description

Technical Field

[0001] This invention relates to the field of information security technology, and in particular to a method for optimizing data transmission in a graphics workstation based on information security. Background Technology

[0002] With the widespread application of high-performance graphics workstations in engineering design, simulation calculation, visualization rendering, and remote collaborative work, the large-scale graphics data generated by these workstations needs to be transmitted over networks at high frequency and low latency. In practical applications, graphics data often has characteristics such as large data volume, high real-time requirements, and strong transmission continuity. It also faces the challenges of complex network environments and diverse security threats, thus placing higher demands on the security and stability of data transmission.

[0003] Existing data transmission technologies for graphics workstations primarily focus on improving network bandwidth utilization or transmission rates. They typically rely on fixed rules or static strategies for path selection and scheduling control, making it difficult to uniformly model and dynamically respond to changes in network status, security risk fluctuations, and changes in the trustworthiness of transmission nodes. In complex network environments, network congestion, link quality fluctuations, and abnormal traffic injection can easily lead to decreased data transmission efficiency, or even data loss or transmission interruption. Furthermore, some existing technologies only improve data transmission security through simple encryption or authentication methods, lacking coordinated control over the overall security constraints of the transmission path, making it difficult to achieve comprehensive optimization of security and efficiency in multi-path transmission scenarios. In addition, existing technologies generally lack the ability to globally characterize the transmission situation, failing to uniformly represent and analyze network status parameters, security risk parameters, and node trustworthiness parameters. This results in transmission control decisions relying on empirical rules or local information, leading to insufficient adaptability and robustness.

[0004] Therefore, how to provide a method for optimizing data transmission in a graphics workstation based on information security is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a data transmission optimization method for graphics workstations based on information security. This invention utilizes transmission situation modeling, enhanced Fourier neural operator models, and improved Decision Transformer models to optimize and control the transmission process of graphics data generated by graphics workstations in complex network and security constraint environments. By constructing a transmission situation feature field, generating a path constraint matrix and a feasible set of transmission control actions, and introducing a projection dual mechanism to achieve collaborative transmission control action generation, it completes multi-path secure scheduling transmission and adaptive updates of the transmission situation. It has the advantages of high transmission security, strong action feasibility, strong adaptability, and high transmission efficiency.

[0006] A method for optimizing data transmission in a graphics workstation based on information security, according to an embodiment of the present invention, includes the following steps: S1. Acquire the graphics data to be transmitted generated by the graphics workstation, and simultaneously collect a set of transmission status parameters related to data transmission. S2. Construct a transmission situation feature field from the set of transmission situation parameters, input it into the enhanced Fourier neural operator model, generate a path constraint structure field and discretize it to obtain a path constraint matrix, and generate a feasible set of transmission control actions based on the path constraint matrix. S3. Jointly encode the network state parameters and security risk parameters in the transmission status parameter set to generate a decision state feature vector sequence. S4. Input the decision state feature vector sequence and the feasible set of transmission control actions into the improved DecisionTransformer model, introduce the projection dual mechanism, and generate cooperative transmission control actions in the action space limited by the feasible set of transmission control actions. S5. Based on the cooperative transmission control action, perform multi-path secure scheduling transmission control on the graphic data to be transmitted to generate a multi-path transmission data stream; S6. Calculate the path constraint margin field based on multi-path transmission data streams, cooperative transmission control actions, and path constraint matrices. S7. Update the transmission situation feature field according to the path constraint margin field to generate an updated transmission situation feature field. The updated transmission situation feature field is used as the input of the new transmission situation feature field into the enhanced Fourier neural operator model.

[0007] Optionally, S1 specifically includes: The system acquires the graphics data to be transmitted from the graphics workstation; it collects a set of transmission status parameters related to data transmission; the set of transmission status parameters includes network status parameters, security risk parameters, and node trust parameters; the network status parameters include link bandwidth parameters, end-to-end latency parameters, packet loss rate parameters, and link congestion parameters; the security risk parameters include intrusion alarm parameters, abnormal traffic characteristic parameters, and data leakage risk parameters; the node trust parameters include transmission node trust level parameters, transmission path trust level parameters, and authentication status parameters. The transmission status parameter set is time-aligned and normalized to form a transmission status parameter set.

[0008] Optionally, the step of constructing a transmission situation feature field from the set of transmission situation parameters, inputting it into an enhanced Fourier neural operator model, generating a path constraint structure field, and discretizing it to obtain a path constraint matrix is ​​specifically as follows: The network state parameters, security risk parameters, and node trust parameters in the transmission situation parameter set are dimensionally aligned according to a preset parameter order to form a situation parameter tensor with a unified dimension. According to the preset spatial partitioning rules, the situation parameter tensor is mapped to the regular grid, and the situation parameter values ​​in each grid cell of the regular grid are aggregated to generate the transmission situation feature field. The transmission situation feature field is input into the enhanced Fourier neural operator model, and the output is the path constraint structure field; The continuous values ​​in the path constraint structure field are mapped to the preset quantization interval to generate the path constraint matrix.

[0009] Optionally, the step of inputting the transmitted situational feature field into the enhanced Fourier neural operator model and outputting a path constraint structure field specifically involves: The transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid; according to the preset dimensional order, a fast Fourier transform operation is performed on the numerical tensor along the spatial dimension to obtain a frequency domain feature representation containing amplitude and phase components. Frequency components within a preset frequency range are selected from the frequency domain feature representation; the selected frequency components are linearly combined according to preset frequency domain mapping parameters to obtain an intermediate frequency domain mapping result; the intermediate frequency domain mapping result is processed element-wise to obtain a frequency domain mapping result. The frequency domain mapping results are arranged into frequency domain tensors according to frequency indices; an inverse Fourier transform operation is performed on the frequency domain tensors along the spatial dimension to obtain a spatial domain feature representation consistent with the spatial structure of the transmission situation feature field. The spatial domain feature representation is used as the input feature matrix. A linear transformation operation is performed on the input feature matrix according to a preset linear mapping parameter to output a path constraint structure field.

[0010] Optionally, S3 specifically includes: The network state parameters and security risk parameters in the transmission situation parameter set are aligned according to a preset parameter order to form a joint parameter vector; For each dimension of the joint parameter vector, linear scaling is performed based on the maximum and minimum values ​​of the corresponding parameter within a preset time window to map the parameter values ​​to a unified numerical range. Scale adjustment is then performed based on the mean and standard deviation of the corresponding parameter. Normalization is then performed according to the parameter type, and the processing results are combined to form a normalized parameter vector. According to the chronological order of the acquisition time of the transmission status parameter set, the normalized parameter vectors are arranged and spliced ​​to form a multidimensional vector sequence; The dimensions of the multidimensional vector sequence are rearranged to form a fixed-dimensional state vector, and the state vectors are combined in chronological order to form a decision state feature vector sequence.

[0011] Optionally, S4 specifically includes: The decision state feature vector sequence is input into the situation semantic encoding module to generate a semantic state sequence; the feasible set of transmission control actions is input into the budget construction module to generate an action budget sequence; the semantic state sequence and the action budget sequence are aligned to form a joint sequence. The joint sequence is input into the feasible projection module, and the action candidate calculation is performed on the joint sequence based on the projection dual mechanism to obtain the action candidate sequence. The feasible domain projection is then performed on the action candidate sequence to generate the cooperative transmission control action. The cooperative transmission control actions and semantic state sequences are input into the policy update module to generate update policy parameters.

[0012] Optionally, the improved Decision Transformer model specifically includes a situational semantic encoding module, a budget construction module, a feasible projection module, and a policy update module; The situational semantic encoding module performs positional encoding on the decision state feature vector sequence in chronological order to form a state vector sequence with time markers; performs linear transformation on the state vector sequence with time markers to obtain a state embedding vector sequence; and performs sequence attention calculation on the state embedding vector sequence to obtain a semantic state sequence. The budget construction module vectorizes the action parameters in the feasible set of transmission control actions to form an action vector set; arranges the action vector set according to a preset constraint order to form an action budget sequence; and performs scale normalization processing on the action budget sequence to obtain a normalized action budget sequence. The feasible projection module introduces a projection dual mechanism, which includes concatenating the semantic state sequence and the action budget sequence to form a joint input sequence; performing action candidate calculation on the joint input sequence to obtain an action candidate sequence; performing dual variable update on the action candidate sequence to obtain a dual constraint vector; and performing projection on the action candidate sequence based on the dual constraint vector to obtain a cooperative transmission control action. The policy update module pairs the collaborative transmission control actions with the semantic state sequences according to the time index; performs sequence modeling processing on the pairing results to generate a policy update vector; and updates the parameters of the improved Decision Transformer model based on the policy update vector.

[0013] Optionally, S5 specifically includes: The graphic data to be transmitted is segmented according to a preset segmentation rule to form a set of graphic data segments. Based on the collaborative transmission control action, the fragment identifier and path identifier in the graphics data fragment set are matched to form the fragment path allocation result; The graphics data fragment set is split and scheduled according to the fragment path allocation result to form a multi-path fragmented data stream. The multipath fragmented data stream is encapsulated according to the transmission parameters in the cooperative transmission control action to generate a multipath transmission data stream.

[0014] Optionally, S6 specifically includes: The path identifier parsing process is performed on the multipath transmission data stream to form a path data stream sequence; the path data stream sequence is aligned with the cooperative transmission control action to form a path action joint sequence. The path action joint sequence is indexed and mapped based on the path index of the path constraint matrix to form a path constraint aligned sequence. Perform a difference operation on the joint vector of path actions in the path constraint alignment sequence and the constraint values ​​in the path constraint matrix to form a path constraint margin sequence; The path constraint margin sequence is mapped to a regular grid according to the path index to form a path constraint margin field.

[0015] Optionally, S7 specifically includes: The path constraint margin field is dimension-aligned to form a margin tensor; the margin tensor is mapped to a regular grid according to a preset spatial partitioning rule to form a margin feature field. The transmission situation feature field and the margin feature field are fused at the element level to generate an updated transmission situation feature field. The updated transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid and input into the enhanced Fourier neural operator model.

[0016] The beneficial effects of this invention are: (1) By constructing a transmission situation feature field and introducing an enhanced Fourier neural operator model, network state parameters, security risk parameters and node trust parameters are modeled in a unified manner, and a path constraint matrix and a feasible set of transmission control actions are generated, so as to realize the global characterization of the transmission situation under complex network and information security constraints, and improve the accuracy and stability of path constraint modeling. (2) An improved Decision Transformer model is introduced in the collaborative transmission control decision-making stage, and combined with the projection dual mechanism, collaborative transmission control actions are generated within the action space limited by the feasible set of transmission control actions. This effectively avoids the generation of infeasible or unsafe transmission actions and improves the executability and safety consistency of transmission control decisions. (3) By calculating the multi-path secure scheduling transmission control and path constraint margin field, the multi-path transmission data stream and the path constraint matrix are dynamically correlated, the remaining path constraint capacity is quantified in real time, and the constraint evaluation results that can be fed back to the transmission control are provided, thereby enhancing the robustness and resource utilization efficiency in the multi-path scheduling process. (4) The transmission situation feature field is adaptively updated based on the path constraint margin field, and the update result is input again into the enhanced Fourier neural operator model to form a closed-loop optimization mechanism, so as to realize the continuous adjustment of the transmission control strategy with the changes in network situation and security status, and improve the adaptive capability and overall transmission efficiency of the graphics workstation data transmission optimization method. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a data transmission optimization method for a graphics workstation based on information security proposed in this invention; Figure 2 This is a schematic diagram of the improved Decision Transformer model proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-2 A method for optimizing data transmission in a graphics workstation based on information security includes the following steps: S1. Acquire the graphics data to be transmitted generated by the graphics workstation, and simultaneously collect a set of transmission status parameters related to data transmission. S2. Construct the transmission situation parameter set into a transmission situation feature field, input it into the enhanced Fourier neural operator model, generate the path constraint structure field and discretize it to obtain the path constraint matrix, and generate a feasible set of transmission control actions based on the path constraint matrix. S3. Jointly encode the network state parameters and security risk parameters in the transmission status parameter set to generate a decision state feature vector sequence. S4. Input the decision state feature vector sequence and the feasible set of transmission control actions into the improved DecisionTransformer model, introduce the projection dual mechanism, and generate cooperative transmission control actions in the action space limited by the feasible set of transmission control actions. S5. Based on the collaborative transmission control action, perform multi-path secure scheduling transmission control on the graphic data to be transmitted to generate a multi-path transmission data stream; S6. Calculate the path constraint margin field based on multi-path transmission data streams, cooperative transmission control actions, and path constraint matrices. S7. Update the transmission situation feature field according to the path constraint margin field to generate an updated transmission situation feature field. The updated transmission situation feature field is used as the input of the new transmission situation feature field into the enhanced Fourier neural operator model.

[0020] In this embodiment, S1 specifically refers to: The graphics data to be transmitted is read from the graphics processing unit and video memory cache of the graphics workstation. The graphics data to be transmitted includes frame buffer data, geometric description data and texture data output by the rendering pipeline, and is formed into a graphics data sequence according to the data generation time order. The system collects a set of transmission status parameters related to data transmission. The network status parameters in the transmission status parameter set are obtained by sampling from the network interface. The link bandwidth parameter is calculated by the ratio of the amount of data successfully sent per unit time to the sampling time length. The end-to-end delay parameter is calculated by the time difference between the sending timestamp and the receiving timestamp. The packet loss rate parameter is calculated by the ratio of the number of lost data packets to the total number of sent data packets. The link congestion parameter is calculated by the ratio of the queuing buffer length to the maximum buffer capacity. Security risk parameters are collected. Intrusion alarm parameters are represented by alarm level values ​​output by the intrusion detection module. Abnormal traffic characteristic parameters are obtained by the ratio of the number of abnormal connections to the total number of connections within a unit of time. Data leakage risk parameters are obtained by the ratio of the number of accesses to sensitive data to the total number of accesses. The trusted parameters of the collection nodes are obtained by comparing the trusted level parameters of the transmission nodes with the ratio of the number of historical successful transmissions to the total number of transmissions. The trusted level parameters of the transmission path are obtained by the weighted average of the trusted level parameters of the nodes on the path. The authentication status parameters are obtained by mapping the authentication results to binary or multi-valued status variables. The transmission status parameter set is time aligned, and parameters with different sampling periods are mapped to a unified time axis through linear interpolation. The linear interpolation method is to calculate the parameter value at the intermediate time according to the time ratio between two adjacent sampling times. The time-aligned transmission status parameter set is normalized, and numerical scaling is performed on each parameter dimension. The scaling method is to subtract the minimum value of the parameter within a preset time window from the current value of the parameter, and then divide by the difference between the maximum and minimum values ​​of the parameter within the preset time window to obtain the normalized parameter value. The normalized network state parameters, security risk parameters, and node trust parameters are arranged in a preset parameter order to form a transmission status parameter set.

[0021] In this embodiment, the transmission situation parameter set is constructed into a transmission situation feature field, and input into an enhanced Fourier neural operator model to generate a path constraint structure field, which is then discretized to obtain the path constraint matrix, specifically: The network state parameters, security risk parameters, and node trust parameters in the transmission situation parameter set are dimensionally aligned according to the preset parameter order. Different parameters are expanded into vectors of the same length in the time dimension and feature dimension. The dimensional alignment method is to fill missing dimensions with zero values ​​or repeat the parameter values ​​of the most recent time, so that all parameter vectors have a consistent length in the feature dimension, forming a situation parameter tensor. The situation parameter tensor is mapped to a regular grid, which is determined by a preset spatial partitioning rule. The spatial partitioning rule includes a path index dimension and a parameter type dimension. The mapping method is to assign the parameter vector in the situation parameter tensor to the corresponding grid position according to the path index. The situation parameter values ​​in each grid cell are aggregated. The aggregation method is to calculate the weighted average of the parameter values ​​in the grid cell. The weight is determined by the importance coefficient of the parameter in the set of transmitted situation parameters, and a transmitted situation feature field is generated. The transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid and input into an enhanced Fourier neural operator model. The enhancement in the enhanced Fourier neural operator model is represented by introducing a residual superposition structure during the frequency domain mapping process. The residual superposition structure is formed by adding the frequency domain mapping output to the corresponding input frequency domain features to form an enhanced frequency domain feature representation. The model processing includes performing a Fourier transform on the transmission situation feature field along the spatial dimension to obtain frequency domain features, selecting frequency components of the frequency domain features according to a preset frequency index, performing a linear transformation on the selected frequency components and adding them to the original frequency domain features, and then performing an inverse Fourier transform on the enhanced frequency domain features to obtain the path constraint structure field. The continuous values ​​in the path constraint structure field are mapped according to the preset quantization interval. The quantization interval is determined by the minimum constraint value and the maximum constraint value. The interval mapping method is to subtract the minimum constraint value from the value in the path constraint structure field, divide it by the difference between the maximum constraint value and the minimum constraint value, multiply it by the quantization level and round it to obtain the discretized value, forming the path constraint matrix.

[0022] In this embodiment, the transmission situation feature field is input into the enhanced Fourier neural operator model, and the output path constraint structure field is as follows: The transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid. The multidimensional numerical tensor includes path index dimension, spatial location dimension and parameter channel dimension. Each value in the tensor represents the comprehensive value of the situation parameters of the corresponding path at the corresponding spatial location. According to the preset dimensional order, the multidimensional numerical tensor is processed by Fast Fourier Transform along the spatial dimension. Fast Fourier Transform converts the spatial domain representation into the frequency domain representation by performing complex exponential basis function expansion on the discrete numerical values ​​in the spatial dimension. Each frequency component in the frequency domain representation consists of a real part and an imaginary part, which correspond to the amplitude component and the phase component, respectively. Select frequency components within a preset frequency range from the frequency domain feature representation. The preset frequency range is limited by the minimum frequency index and the maximum frequency index. The frequency component selection method is to retain the frequency components whose indices fall within the range and discard the frequency components outside the range. The selected frequency domain components are linearly combined according to the preset frequency domain mapping parameters. The linear combination method is to multiply the selected frequency domain components with the corresponding mapping weights and sum them in the frequency dimension to form an intermediate frequency domain mapping result. The intermediate frequency domain mapping result is processed element-wise. The element-wise operation method is to sum each frequency domain component in the intermediate frequency domain mapping result with the corresponding input frequency domain component to form the enhanced frequency domain mapping result. The enhancement means that the frequency domain mapping result contains the original frequency domain information and the mapped frequency domain information. The frequency domain mapping results are arranged into a frequency domain tensor according to the frequency index order, and the frequency index in the frequency domain tensor is consistent with the spatial index. The frequency domain tensor is processed by inverse Fourier transform along the spatial dimension. The inverse Fourier transform restores the frequency domain representation to a spatial domain feature representation consistent with the spatial structure of the transmission situation feature field by performing complex exponential basis function superposition on the frequency components in the frequency domain tensor. The spatial domain feature representation is used as the input feature matrix. The input feature matrix is ​​then subjected to a linear transformation according to a preset linear mapping parameter. The linear transformation method is to multiply the input feature matrix with the mapping weight matrix and then superimpose the bias vector to obtain the path constraint structure field.

[0023] In this embodiment, S3 specifically refers to: The network status parameters and security risk parameters in the transmission status parameter set are aligned according to a preset parameter order. The preset parameter order is arranged in the order of link bandwidth parameters, end-to-end delay parameters, packet loss rate parameters, link congestion parameters, intrusion alarm parameters, abnormal traffic characteristic parameters, and data leakage risk parameters. The alignment method is to arrange the corresponding parameters in sequence under the same time index to form a joint parameter vector. The parameter in each dimension of the joint parameter vector is linearly scaled. The linear scaling method is to subtract the minimum value of the parameter within a preset time window from the current value of the parameter, and then divide by the difference between the maximum and minimum values ​​of the parameter within the preset time window, so that the parameter value is mapped to a uniform numerical range of 0 to 1. After linear scaling, the parameter is scaled. The scaling method is to subtract the mean of the parameter within a preset time window from the parameter value, and then divide it by the standard deviation of the parameter within the preset time window. The standard deviation is obtained by taking the square root of the average of the squares of the differences between the parameter value and the mean. Normalization is performed according to parameter type. The scaling results corresponding to network state parameters and security risk parameters are arranged and concatenated according to the preset parameter order. The concatenation method is to connect the first and last vectors in the feature dimension to form a normalized parameter vector. According to the chronological order of the acquisition time of the transmission status parameter set, the normalized parameter vectors under the continuous time index are arranged and spliced. The splicing method is to connect multiple normalized parameter vectors in sequence along the time dimension to form a multi-dimensional vector sequence. The multidimensional vector sequence is rearranged in a dimension-rearranged manner by rearranging the time dimension and feature dimension into a fixed-length vector representation, so that each time index corresponds to a fixed-dimensional state vector. The state vectors are combined in chronological order to form a sequence of decision state feature vectors.

[0024] In this embodiment, S4 specifically refers to: The decision state feature vector sequence is input into the situation semantic encoding module to generate a semantic state sequence; the feasible set of transmission control actions is input into the budget construction module to generate an action budget sequence; the semantic state sequence and the action budget sequence are aligned to form a joint sequence. The joint sequence is input into the feasible projection module, and the action candidate calculation is performed on the joint sequence based on the projection dual mechanism to obtain the action candidate sequence. The feasible domain projection is then performed on the action candidate sequence to generate the cooperative transmission control action. The cooperative transmission control actions and semantic state sequences are input into the policy update module to generate update policy parameters.

[0025] In this embodiment, the improved Decision Transformer model specifically includes a situational semantic encoding module, a budget construction module, a feasible projection module, and a policy update module; The situational semantic encoding module performs positional encoding on the decision state feature vector sequence according to the time index. Positional encoding involves mapping the time index to a position vector with the same dimension as the decision state feature vector, and then adding it element-wise to the corresponding decision state feature vector in the feature dimension to form a state vector sequence with time labels. The state vector sequence with time labels is then subjected to linear transformation, which involves multiplying the state vector with the state mapping weight matrix and superimposing the state bias vector to obtain a state embedding vector sequence. The state embedding vector sequence is then subjected to sequence attention calculation, which involves calculating the similarity between any two state embedding vectors in the time dimension. The similarity is obtained through vector dot product, and the similarity result is normalized. The state embedding vectors are then weighted and summed based on the normalized weight values ​​to form a semantic state sequence. Normalization specifically involves: performing exponential mapping on the similarity results obtained through the dot product of any two state embedding vectors in the time dimension; summing the exponential mapping results obtained under the same time index to obtain the exponential mapping sum; dividing each exponential mapping result by the exponential mapping sum under the corresponding time index to obtain the normalized weight value; the normalized weight value satisfies that its value ranges from 0 to 1 under the same time index, and the sum of the normalized weight values ​​is 1; and performing a weighted summation on the corresponding state embedding vectors based on the normalized weight values. The budget construction module vectorizes the action parameters in the feasible set of transmission control actions. Vectorization involves arranging the parameters of each transmission control action according to the path index and parameter dimension order to form an action vector set. The action vector set is then arranged according to a preset constraint order, which is sorted from small to large path index and from low to high action parameter dimension to form an action budget sequence. The action budget sequence is then scaled and normalized. Scale normalization involves subtracting the minimum value of the corresponding parameter in the action budget sequence from each parameter value in the action budget sequence, and then dividing by the difference between the maximum and minimum values ​​of the corresponding parameter in the action budget sequence to obtain a normalized action budget sequence. The feasible projection module introduces a projection dual-duality mechanism, which concatenates the semantic state sequence and the action budget sequence. This concatenation involves placing the semantic state vector at the beginning and the action budget vector at the end along the feature dimension, forming a joint input sequence. The joint input sequence is then processed to calculate action candidates. This involves multiplying the vectors in the joint input sequence by the action generation weight matrix and superimposing the action bias vector to form a candidate action sequence. The candidate action sequence is then updated with dual-duality variables. This involves calculating the deviation between each action component in the candidate action sequence and the corresponding boundary value of the feasible set of transmission control actions, and mapping the deviation to a dual constraint vector. Finally, the candidate action sequence is projected based on the dual constraint vector. This projection process trims components in the candidate action sequence that exceed the boundary of the feasible set of transmission control actions to the corresponding boundary value range, generating cooperative transmission control actions. The policy update module pairs cooperative transmission control actions and semantic state sequences according to their time indices. The pairing method involves concatenating the cooperative transmission control action vector and the semantic state vector under the same time index along the feature dimension. Sequence modeling is then performed on the concatenated vector, which involves linearly mapping and weighted accumulation of the concatenated vector along the time dimension to form the policy update vector. The parameters in the improved DecisionTransformer model are updated based on the policy update vector by applying incremental adjustments to the model parameters according to the policy update vector.

[0026] In this implementation, existing Decision Transformer models primarily rely on historical states and action sequences for sequence modeling, lacking explicit characterization of the feasible set of transmission control actions and path constraint information. This makes it difficult to ensure the feasibility and security consistency of generated actions under complex information security constraints and multi-path transmission scenarios. Therefore, this implementation introduces the feasible set of transmission control actions into the sequence modeling framework of the Decision Transformer model to participate in decision modeling. By jointly representing the decision state feature vector sequence and the action budget sequence, the model can perceive the feasible boundary of the action space while understanding the state semantics. Furthermore, a projection dual mechanism is introduced in the action generation stage to transform the path constraint matrix and the feasible set of transmission control actions into dual constraint information. The action candidate results are then constrained and corrected through dual variable updates and feasible domain projection processing, ensuring that the generated collaborative transmission control actions always remain within the feasible action space. Through these improvements, the model maintains the advantages of Decision Transformer sequence decision-making while achieving collaborative modeling of information security constraints, multi-path resource limitations, and transmission situation changes. This results in a transmission control decision-making capability that balances security and executability, demonstrating a distinct advantage over existing Decision Transformer models in the context of data transmission optimization on graphical workstations. Structural and decision-making mechanism innovations in the Transformer model.

[0027] In this embodiment, S5 specifically refers to: The graphic data to be transmitted is fragmented according to a preset fragmentation rule. The preset fragmentation rule is to divide the data according to the data byte length and frame boundary, so that each fragment contains a continuous data byte range. The fragment byte length is determined by the fragmentation size parameter in the cooperative transmission control action, forming a set of graphic data fragments. Each piece in the set of graphics data pieces is assigned a unique piece identifier, which is determined by the starting byte position of the piece in the original graphics data and the piece sequence number. Based on the cooperative transmission control action, the fragment identifier and path identifier in the graphics data fragment set are matched. The matching method is to map each fragment identifier to the corresponding path identifier according to the path selection parameters given in the cooperative transmission control action. The path identifier corresponds to the transmission path index limited in the feasible set of transmission control actions, forming the fragment path allocation result. The graphics data fragment set is split and scheduled according to the fragment path allocation result. The split and scheduling process is to divide the fragment set into multiple subsets based on the path identifier. Each subset corresponds to a transmission path, forming a multi-path fragmented data stream. The multipath fragmented data stream is encapsulated according to the transmission parameters in the cooperative transmission control action. The encapsulation process involves adding a path identifier field, a fragment sequence number field, and a check field to the front of each fragment. The check field is obtained by summing the fragment byte values ​​bit by bit and taking the modulo, thus generating a multipath transmission data stream.

[0028] In this embodiment, S6 specifically refers to: The multipath transmission data stream is processed by path identifier parsing. The path identifier parsing process involves reading the path identifier field from the fragment encapsulation header of each fragment in the multipath transmission data stream, and classifying and reorganizing the fragment data according to the path identifier field to form a path data stream sequence sorted by path index. Alignment processing is performed on the path data stream sequence and the cooperative transmission control action. The alignment processing is to associate the path data stream corresponding to the same path in the same time slice with the corresponding action vector in the cooperative transmission control action according to the time index and path index, and then concatenate them in the feature dimension. The concatenation method is to place the path data stream statistical features at the beginning and the action vector at the end and connect them end to end to form a path action joint sequence. The path action joint sequence is indexed and mapped according to the path index of the path constraint matrix. The index mapping process is to read the constraint value vector of the corresponding path from the path constraint matrix according to the path index and align it with the corresponding joint vector in the path action joint sequence on the time index to form a path constraint alignment sequence. The path action joint vector in the path constraint alignment sequence is subjected to a difference operation with the constraint value in the path constraint matrix. The difference operation is to subtract the corresponding path constraint value from the component in the path action joint vector that represents resource occupation or transmission intensity. A positive difference result indicates the remaining amount of path constraint, and a negative difference result indicates the excess amount of path constraint, thus forming a path constraint margin sequence. The path constraint margin sequence is mapped to a regular grid according to the path index. The horizontal axis of the regular grid corresponds to the path index, and the vertical axis corresponds to the time index. The grid cell values ​​are filled by the path constraint margin values ​​under the corresponding time index and path index, forming a path constraint margin field.

[0029] In this embodiment, S7 specifically refers to: The path constraint margin field is dimensionally aligned. The dimensional alignment process involves mapping the path index dimension and time index dimension in the path constraint margin field to the path index dimension and spatial location dimension corresponding to the transmission situation feature field. Missing time index positions are filled by copying the path constraint margin value under the adjacent time index, forming a margin tensor that is consistent with the transmission situation feature field in terms of dimensional structure. The margin tensor is mapped to a regular grid according to a preset spatial partitioning rule. The preset spatial partitioning rule is consistent with the regular grid used when constructing the transmission situation feature field. The mapping method is to fill the corresponding grid cells of the regular grid with the values ​​of the corresponding path index and spatial location index in the margin tensor to form the margin feature field. The transmission situation feature field and the margin feature field are fused at the element level. The element level fusion process is to perform a weighted summation of the transmission situation feature field values ​​and the margin feature field values ​​located at the same grid position. The weighting method is to multiply the transmission situation feature field value by a preset situation weight coefficient and the margin feature field value by a preset margin weight coefficient, and then add them together. The sum of the situation weight coefficient and the margin weight coefficient is one, thereby generating an updated transmission situation feature field. The updated transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid. Each value in the multidimensional numerical tensor corresponds to the fusion result under the path index, spatial location index and parameter channel index. The multidimensional numerical tensor is then used as the input feature to input the enhanced Fourier neural operator model again.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a 3D visualization remote collaboration scenario in an engineering design institute. The graphics workstation was deployed in the intranet rendering area, and end users were distributed across two network environments: the campus office network and external encrypted leased lines. The business content included real-time rotation of the 3D assembly model, material switching, and section browsing. The graphics data to be transmitted was continuously generated by the graphics workstation, with single-frame data volume fluctuating between 80MB and 260MB, and a peak frame rate of 30fps. During collaboration, intrusion alarms and abnormal traffic surges occurred simultaneously. Network link bandwidth jittered between 120Mbps and 900Mbps, end-to-end latency jittered between 18ms and 210ms, packet loss rate varied between 0.05% and 3.20%, and link congestion parameters varied between 0.08 and 0.92. Traditional static multipath strategies tend to allocate high-security-risk links to high-occupancy data streams when abnormal traffic occurs, leading to concurrent transmission interruptions and frame stuttering, creating a problem where real-time performance and security are difficult to balance. In this scenario, the present invention continuously collects a set of transmission situation parameters after the session is established. This set includes network state parameters, security risk parameters, and node trust parameters, which are then time-aligned and normalized to obtain a transmission situation parameter set that can be directly used for modeling. The transmission situation parameter set is used to construct a transmission situation feature field, which is then input into an enhanced Fourier neural operator model. The output path constraint structure field is discretized to obtain a path constraint matrix. Simultaneously, a feasible set of transmission control actions is generated based on the path constraint matrix. The network state parameters and security risk parameters are jointly encoded to form a decision state feature vector sequence. This sequence, along with the feasible set of transmission control actions, is input into an improved Decision class. The Transformer model introduces a projective dual mechanism to generate cooperative transmission control actions within the action space defined by the feasible set of transmission control actions. These cooperative transmission control actions drive multi-path secure scheduling transmission control, fragmenting the graphics data to be transmitted according to fragment size parameters to form a set of graphics data fragments. The fragment identifier is determined by the starting byte position and fragment sequence number. Matching the fragment identifier with the path identifier forms the fragment path allocation result. After encapsulation, the multi-path fragmented data stream forms a multi-path transmission data stream. The multi-path transmission data stream, cooperative transmission control actions, and path constraint matrix jointly participate in the calculation of the path constraint margin field. The path constraint margin field is fed back to update the transmission situation feature field, generating an updated transmission situation feature field, which is then input again into the enhanced Fourier neural operator model, forming a closed-loop scheduling mechanism oriented towards security constraints and network jitter. To quantify the beneficial effects of the present invention, the "fixed weight multipath scheduling" and the "method of the present invention" were compared under the same business load. The stable period for two consecutive hours and the disturbance period containing two intrusion alarms and one abnormal traffic peak were statistically analyzed, resulting in Table 1.

[0031] Table 1. Comparison of Transmission Performance and Security in Remote Collaboration Scenarios

[0032] Table 1 shows that the method of the present invention reduces the end-to-end latency P95 from 96ms to 61ms in the stable phase, while increasing the effective throughput to 486Mbps. This indicates that the cooperative transmission control action can make better use of the low-congestion link under the constraint of the feasible set of transmission control actions. In the disturbance phase, the method of the present invention still controls the packet loss rate at 0.74% and the fragmentation retransmission rate at 1.85% under intrusion alarms and abnormal traffic peaks. The number of frame stutters is reduced from 29 times per hour to 6 times per hour, and the duration of the impact after the intrusion alarm is shortened to 23s. This shows that the path constraint matrix and projection dual mechanism have a direct effect on the suppression of high-risk paths and the correction of infeasible actions by feasible domain projection.

[0033] To further verify the contribution of closed-loop updates, the path constraint margin field and transmission status characteristic field were compared before and after the update. Four typical transmission paths were selected, and the average path constraint margin value, number of overruns and link utilization were statistically analyzed every 10 minutes during the disturbance segment, resulting in Table 2.

[0034] Table 2 Comparison of Path Constraint Margin Field and Scheduling Results for Disturbance Segment

[0035] In Table 2, the fixed-weight multipath scheduling shows negative average path constraint margin values ​​and a high number of overruns in paths 1, 2, and 4, indicating that scheduling actions frequently trigger the constraint boundaries corresponding to the path constraint matrix. The method of this invention pulls the average path constraint margin values ​​of all four paths back to the positive range, while reducing the number of overruns to 0 to 2 times per 10 minutes. This shows that the quantification of the constraint remaining amount by the path constraint margin field can effectively drive the update of the transmission situation feature field, making the path constraint structure field output by the enhanced Fourier neural operator model more closely match the disturbance potential. The coordinated transmission control actions fall more stably into the action space under the action of the projection dual mechanism, thereby improving the link utilization of paths 1, 2, and 3 and reducing the risk of overruns without increasing interruption.

[0036] Under the condition of complex network jitter and security risk fluctuations, this invention achieves secure and controllable scheduling of multi-path transmission data streams through the collaborative closed loop of transmission situation feature field, path constraint matrix, feasible set of transmission control actions and projection dual mechanism. This significantly reduces the interactive stuttering caused by high latency and high retransmission, while shortening the duration of the impact of security events on services and improving overall throughput efficiency.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing data transmission in a graphics workstation based on information security, characterized in that, Includes the following steps: S1. Acquire the graphics data to be transmitted generated by the graphics workstation, and simultaneously collect a set of transmission status parameters related to data transmission. S2. Construct a transmission situation feature field from the set of transmission situation parameters, input it into the enhanced Fourier neural operator model, generate a path constraint structure field and discretize it to obtain a path constraint matrix, and generate a feasible set of transmission control actions based on the path constraint matrix. S3. Jointly encode the network state parameters and security risk parameters in the transmission status parameter set to generate a decision state feature vector sequence. S4. Input the decision state feature vector sequence and the feasible set of transmission control actions into the improved DecisionTransformer model, introduce the projection dual mechanism, and generate cooperative transmission control actions in the action space limited by the feasible set of transmission control actions. S5. Based on the cooperative transmission control action, perform multi-path secure scheduling transmission control on the graphic data to be transmitted to generate a multi-path transmission data stream; S6. Calculate the path constraint margin field based on multi-path transmission data streams, cooperative transmission control actions, and path constraint matrices. S7. Update the transmission situation feature field according to the path constraint margin field to generate an updated transmission situation feature field. The updated transmission situation feature field is used as the input of the new transmission situation feature field into the enhanced Fourier neural operator model.

2. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, Specifically, S1 is: The system acquires the graphics data to be transmitted from the graphics workstation; it collects a set of transmission status parameters related to data transmission; the set of transmission status parameters includes network status parameters, security risk parameters, and node trust parameters; the network status parameters include link bandwidth parameters, end-to-end latency parameters, packet loss rate parameters, and link congestion parameters; the security risk parameters include intrusion alarm parameters, abnormal traffic characteristic parameters, and data leakage risk parameters; the node trust parameters include transmission node trust level parameters, transmission path trust level parameters, and authentication status parameters. The transmission status parameter set is time-aligned and normalized to form a transmission status parameter set.

3. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, The process of constructing a transmission situation feature field from the set of transmission situation parameters, inputting it into an enhanced Fourier neural operator model, generating a path constraint structure field, and discretizing it to obtain the path constraint matrix is ​​as follows: The network state parameters, security risk parameters, and node trust parameters in the transmission situation parameter set are dimensionally aligned according to a preset parameter order to form a situation parameter tensor with a unified dimension. According to the preset spatial partitioning rules, the situation parameter tensor is mapped to the regular grid, and the situation parameter values ​​in each grid cell of the regular grid are aggregated to generate the transmission situation feature field. The transmission situation feature field is input into the enhanced Fourier neural operator model, and the output is the path constraint structure field; The continuous values ​​in the path constraint structure field are mapped to the preset quantization interval to generate the path constraint matrix.

4. The method for optimizing data transmission in a graphics workstation based on information security according to claim 3, characterized in that, The process of inputting the transmission situation feature field into the enhanced Fourier neural operator model and outputting the path constraint structure field is specifically as follows: The transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid; according to the preset dimensional order, a fast Fourier transform operation is performed on the numerical tensor along the spatial dimension to obtain a frequency domain feature representation containing amplitude and phase components. Select frequency domain components within a preset frequency range from the frequency domain feature representation; The selected frequency domain components are linearly combined according to preset frequency domain mapping parameters to obtain an intermediate frequency domain mapping result; the intermediate frequency domain mapping result is then processed element-wise to obtain the frequency domain mapping result. The frequency domain mapping results are arranged into frequency domain tensors according to frequency indices; an inverse Fourier transform operation is performed on the frequency domain tensors along the spatial dimension to obtain a spatial domain feature representation consistent with the spatial structure of the transmission situation feature field. The spatial domain feature representation is used as the input feature matrix. A linear transformation operation is performed on the input feature matrix according to a preset linear mapping parameter to output a path constraint structure field.

5. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, Specifically, S3 is: The network state parameters and security risk parameters in the transmission situation parameter set are aligned according to a preset parameter order to form a joint parameter vector; For each dimension of the joint parameter vector, linear scaling is performed based on the maximum and minimum values ​​of the corresponding parameter within a preset time window to map the parameter values ​​to a unified numerical range. Scale adjustment is then performed based on the mean and standard deviation of the corresponding parameter. Normalization is then performed according to the parameter type, and the processing results are combined to form a normalized parameter vector. According to the chronological order of the acquisition time of the transmission status parameter set, the normalized parameter vectors are arranged and spliced ​​to form a multidimensional vector sequence; The dimensions of the multidimensional vector sequence are rearranged to form a fixed-dimensional state vector, and the state vectors are combined in chronological order to form a decision state feature vector sequence.

6. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, Specifically, S4 is: The decision state feature vector sequence is input into the situation semantic encoding module to generate a semantic state sequence; the feasible set of transmission control actions is input into the budget construction module to generate an action budget sequence; the semantic state sequence and the action budget sequence are aligned to form a joint sequence. The joint sequence is input into the feasible projection module, and the action candidate calculation is performed on the joint sequence based on the projection dual mechanism to obtain the action candidate sequence. The feasible domain projection is then performed on the action candidate sequence to generate the cooperative transmission control action. The cooperative transmission control actions and semantic state sequences are input into the policy update module to generate update policy parameters.

7. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, The improved Decision Transformer model specifically includes a situational semantic encoding module, a budget construction module, a feasible projection module, and a policy update module; The situational semantic encoding module performs positional encoding on the decision state feature vector sequence in chronological order to form a state vector sequence with time markers; performs linear transformation on the state vector sequence with time markers to obtain a state embedding vector sequence; and performs sequence attention calculation on the state embedding vector sequence to obtain a semantic state sequence. The budget construction module vectorizes the action parameters in the feasible set of transmission control actions to form an action vector set; arranges the action vector set according to a preset constraint order to form an action budget sequence; and performs scale normalization processing on the action budget sequence to obtain a normalized action budget sequence. The feasible projection module introduces a projection dual mechanism, which includes concatenating the semantic state sequence and the action budget sequence to form a joint input sequence; performing action candidate calculation on the joint input sequence to obtain an action candidate sequence; performing dual variable update on the action candidate sequence to obtain a dual constraint vector; and performing projection on the action candidate sequence based on the dual constraint vector to obtain a cooperative transmission control action. The policy update module pairs the coordinated transmission control actions with the semantic state sequences according to the time index; Sequence modeling is performed on the pairing results to generate a policy update vector; the parameters of the improved DecisionTransformer model are then updated based on the policy update vector.

8. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, Specifically, S5 is: The graphic data to be transmitted is segmented according to a preset segmentation rule to form a set of graphic data segments. Based on the collaborative transmission control action, the fragment identifier and path identifier in the graphics data fragment set are matched to form the fragment path allocation result; The graphics data fragment set is split and scheduled according to the fragment path allocation result to form a multi-path fragmented data stream. The multipath fragmented data stream is encapsulated according to the transmission parameters in the cooperative transmission control action to generate a multipath transmission data stream.

9. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, Specifically, S6 is: The path identifier parsing process is performed on the multipath transmission data stream to form a path data stream sequence; the path data stream sequence is aligned with the cooperative transmission control action to form a path action joint sequence. The path action joint sequence is indexed and mapped based on the path index of the path constraint matrix to form a path constraint aligned sequence. Perform a difference operation on the joint vector of path actions in the path constraint alignment sequence and the constraint values ​​in the path constraint matrix to form a path constraint margin sequence; The path constraint margin sequence is mapped to a regular grid according to the path index to form a path constraint margin field.

10. The method for optimizing data transmission in a graphics workstation based on information security according to claim 1, characterized in that, Specifically, S7 is: The path constraint margin field is dimension-aligned to form a margin tensor; the margin tensor is mapped to a regular grid according to a preset spatial partitioning rule to form a margin feature field. The transmission situation feature field and the margin feature field are fused at the element level to generate an updated transmission situation feature field. The updated transmission situation feature field is represented as a multidimensional numerical tensor on a regular grid and input into the enhanced Fourier neural operator model.

Images