An intelligent game and operational research optimization assisted decision-making system and method with atlas enhancement

Abstract

The present invention relates to the technical field of auxiliary decision-making, and discloses a graph-enhanced intelligent game and operational research optimization auxiliary decision-making system and method. The graph-enhanced intelligent game and operational research optimization auxiliary decision-making method includes: inputting multi-source heterogeneous data into a spatio-temporal semantic mapping framework to generate a unified spatio-temporal semantic data structure; generating a cognitive-spatio-temporal manifold model through a cognitive spatio-temporal manifold generation algorithm; obtaining user interaction data and generating an adaptive view expression through an interaction response manifold deformation algorithm; obtaining user interaction behavior data and generating a cognitive balance view through a cognitive load perception projection algorithm; generating an interactive decision-making scene through an immersive spatio-temporal scene rendering technology. By mapping complex game and operational research optimization problems into intuitive spatio-temporal scenes, the present invention enables decision-makers to foresee and manipulate decision-making results in a virtual spatio-temporal environment, thereby greatly enhancing the understandability and controllability of games and operational research optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of auxiliary decision-making, and more specifically, it relates to an intelligent game and operational research optimization auxiliary decision-making system and method enhanced by a knowledge graph. Background Art

[0002] In the field of complex game and operational research optimization decision-making, there are significant technical bottlenecks in existing decision support systems. Traditional methods mainly use two-dimensional charts and statistical panels for data presentation, and their technical limitations are reflected in:

[0003] Dilemma of high-dimensional abstract data expression: Existing visualization technologies are difficult to effectively process complex decision-making data with more than three dimensions, resulting in decision-makers being unable to intuitively understand the non-linear relationships between multi-dimensional data. Experiments show that when the data dimension exceeds 7, the decision-making understanding efficiency of traditional systems drops by 73%;

[0004] Problem of time dimension fragmentation: Existing systems use discrete time slices to present decision-making impacts and cannot establish a continuous correlation model of multiple time scales. Statistics show that this fragmented presentation increases the probability that decision-makers ignore long-term impacts by 58%;

[0005] Cognitive bias in collaborative decision-making: Traditional multi-user systems only achieve data sharing and lack a cognitive alignment mechanism. Research shows that cognitive differences among decision-makers with different professional backgrounds can lead to 42% of the effective information being lost during the collaboration process;

[0006] Paradox of information loss in dimensionality reduction: Existing visualization methods generally use linear dimensionality reduction techniques such as PCA to adapt to human cognitive abilities, but this will lead to the loss of key decision-making features. Test data shows that traditional methods lose an average of 36% of the decision-related features during the dimensionality reduction process.

[0007] In modern military command and control systems, commanders need to simultaneously process multi-dimensional information such as terrain, meteorology, unit status, logistics supply, and enemy situation, and make decisions in an environment where the battlefield situation changes rapidly. Traditional combat simulation systems usually display this information on multiple interfaces, resulting in commanders having to frequently switch perspectives and making it difficult to form a unified battlefield perception. Research shows that in a high-intensity combat simulation environment, commanders need to switch interfaces 8.3 times per minute on average, and 41% of the key information is ignored during the perspective switching process.

[0008] In the fields of e-sports and strategy games, high-level players need to process dozens of variables such as unit attributes, resource allocation, map control, and skill cooldowns and make responses at the millisecond level. Existing game interface designs mainly rely on two-dimensional plane layouts and linear time axes, and players need to form a mental model through long-term experience. Data shows that even professional players can only process about 42% of the key strategic information simultaneously in a high-pressure environment, resulting in a significant decline in decision-making quality.

[0009] In scenarios such as large-scale disaster emergency response and multinational enterprise strategic planning, decision-makers from different professional fields need to jointly participate in complex decision-making. Due to differences in the professional terms, thinking patterns, and focus of attention they use, traditional systems cannot provide effective cognitive alignment support. Practical case analysis shows that in cross-domain collaborative decision-making, approximately 67% of the communication time is spent on clarifying basic concepts and explaining professional perspectives, rather than discussing the decision itself.

[0010] These technical deficiencies make it difficult for existing systems to meet the triangular requirements of "high-dimensional data understanding - multi-time scale deduction - group cognitive collaboration" in modern complex decision-making scenarios, seriously restricting decision-making quality and efficiency. Summary of the Invention

[0011] The present invention provides a graph-enhanced intelligent game and operational research optimization-assisted decision-making system and method to solve the technical problems in the above-related technologies.

[0012] The present invention provides a graph-enhanced intelligent game and operational research optimization-assisted decision-making method, including the following steps:

[0013] Obtain multi-source heterogeneous data and input it into the spatio-temporal semantic mapping framework to generate a unified spatio-temporal semantic data structure;

[0014] The unified spatio-temporal semantic data structure generates a cognitive-spatio-temporal manifold model through the cognitive spatio-temporal manifold generation algorithm;

[0015] Obtain user interaction data and the cognitive-spatio-temporal manifold model, and generate an adaptive view expression through the interaction response manifold deformation algorithm;

[0016] Obtain user interaction behavior data and the adaptive view expression, and generate a cognitive balance view through the cognitive load perception projection algorithm;

[0017] Generate an interactive decision-making scenario and a multi-modal interaction interface through immersive spatio-temporal scene rendering technology.

[0018] Further, the spatio-temporal semantic mapping framework uses a multi-head self-attention mechanism to achieve cross-domain semantic feature extraction and alignment:

[0019]

[0020] Where represents the aligned data, represents the source data, , and respectively represent semantic descriptors of the time dimension, space dimension, and action dimension, is the semantic alignment function.

[0021] Furthermore, the interactive response manifold deformation algorithm calculates the manifold deformation process through the following partial differential equation:

[0022]

[0023] where represents the user's attention distribution at time t, represents the decision-making operation executed by the user, represents the attention response operator, represents the decision response operator, represents the cognitive-spatiotemporal manifold.

[0024] Furthermore, the interactive response manifold deformation algorithm uses an improved implicit Euler method for numerical solution:

[0025]

[0026] where represents the manifold state at time t, represents the manifold state at time and is the time step.

[0027] Furthermore, the cognitive load perception projection algorithm is implemented through the following steps: Analyze the user interaction behavior data to calculate the cognitive load level value:

[0028]

[0029] where represents the calculated cognitive load level value, represents the user's historical interaction records, represents the user's fixation pattern data, represents the user's operation response time data, is the cognitive load evaluation function;

[0030] Determine the dimension and method of data projection according to the cognitive load level value:

[0031]

[0032] where represents the original high-dimensional data, represents the current decision context information, represents the adaptive projection result, is the adaptive projection function;

[0033] Generate a semantic interpretation for the projection result:

[0034]

[0035] Among them represents the original data, represents the generated projection interpretation text, is the interpretation generation function.

[0036] Furthermore, the cognitive load-aware projection algorithm automatically selects an appropriate dimensionality reduction method according to the cognitive load value:

[0037] When is true, a non-linear dimensionality reduction method that retains the most information is adopted;

[0038] When is true, a dimensionality reduction method that balances information retention and computational efficiency is adopted;

[0039] When is true, a linear dimensionality reduction method with the simplest calculation is adopted;

[0040] Among them represents the low cognitive load threshold, represents the high cognitive load threshold, represents the cognitive load value.

[0041] Furthermore, the immersive spatio-temporal scene rendering technology is realized through the following steps: performing spatio-temporal scene mapping to convert the cognitive balance view into a visual space representation:

[0042]

[0043] Among them represents the visual style parameter set, represents the user preference configuration, is the scene mapping function, represents the visual scene data, represents the cognitive-spatio-temporal manifold model;

[0044] Based on the visual space representation, perform interactive scene rendering to generate an interactive scene:

[0045]

[0046] Among them represents the interactive parameter configuration, represents the physical dynamics parameter set, represents the generated interactive scene, is the scene rendering function.

[0047] Furthermore, after generating the cognitive balance view, it also includes:

[0048] Obtain the interaction data of multiple decision-makers and the cognitive balance view, and generate a collective cognitive structure and a cognitive guidance strategy through a multi-agent cognitive fusion algorithm;

[0049] Provide the collective cognitive structure and the cognitive guidance strategy as additional inputs to the immersive spatio-temporal scene rendering technology to enhance the collaborative decision-making ability of the interactive decision-making scene.

[0050] Furthermore, the multi-agent cognitive fusion algorithm is implemented through the following steps:

[0051] Extract the individual cognitive model from the interaction data of each decision-maker:

[0052]

[0053] Among them, represents the constructed i-th user cognitive model, represents the interaction history data of the i-th user, represents the attention distribution data of the i-th user, represents the decision-making behavior data of the i-th user, is the cognitive model extraction function;

[0054] Fuse multiple individual cognitive models into a collective cognitive model through a graph neural network algorithm:

[0055]

[0056] Among them, represents the graph neural network function, represents the adjacency matrix of the relationship between models, which characterizes the complementary or conflicting relationship between different individual cognitive models, represents starting from the index to the set of individual cognitive models, represents the total number of decision-makers, represents the collective cognitive model;

[0057] Based on the collective cognitive model, identify potential blind spots in group cognition:

[0058]

[0059] Among them, represents the set of identified cognitive blind spots, represents the data representation of the complete decision space, is the detection threshold parameter, represents the blind spot detection function;

[0060] Generate a guidance strategy for the identified cognitive blind spots:

[0061]

[0062] Among them, represents the generated set of cognitive guidance strategies, represents the set of cognitive blind spots, represents the current decision context, represents the strategy generation function.

[0063] A graph-enhanced intelligent game and operational research optimization assisted decision-making system, which is used to execute the above-mentioned graph-enhanced intelligent game and operational research optimization assisted decision-making method, including:

[0064] A spatio-temporal semantic mapping module, which is used to convert multi-source heterogeneous data into a unified spatio-temporal semantic data structure;

[0065] A cognitive manifold generation module, which is used to convert the unified spatio-temporal semantic data structure into a cognitive-spatio-temporal manifold model;

[0066] A manifold deformation module, which is used to adjust the cognitive-spatio-temporal manifold model according to user interaction data and generate an adaptive view expression;

[0067] A cognitive load perception module, which is used to analyze user interaction behavior data, dynamically adjust the dimension and details of data projection, and generate a cognitive balance view;

[0068] A spatio-temporal scene rendering module, which is used to convert the cognitive balance view into an interactive decision-making scene;

[0069] A multi-modal interaction module, which is used to provide gesture interaction, voice interaction and collaborative interaction functions;

[0070] A decision simulation module, which is used to support users to plan decision-making paths and conduct simulations in the scene.

[0071] The beneficial effects of the present invention are as follows:

[0072] Cognitive-spatio-temporal manifold mapping technology: Innovatively propose to map game and operational research optimization problems from an abstract mathematical space to a cognitive-spatio-temporal manifold, making abstract data concrete into a perceivable and operable spatio-temporal scene. This mapping process is different from traditional data visualization technologies. It not only considers the geometric characteristics of data, but also incorporates human cognitive laws, realizing a paradigm shift from "mathematical expression" to "cognitive expression".

[0073] Interaction-induced manifold deformation mechanism: Breakthroughly realizes the two-way coupling of human cognition and data expression. In traditional visualization systems, the data expression form is preset and static, while in this embodiment, the data expression will dynamically deform according to the user's cognitive focus and interaction behavior, realizing the real-time adaptation of data expression and cognitive process.

[0074] Cognitive Adaptive Projection Algorithm: Innovatively solves the contradiction between "data complexity" and "cognitive comprehensibility". Different from traditional methods that simplify data to improve comprehensibility, this algorithm retains all the complexity of the data but dynamically adjusts the dimensions and details of data presentation according to the cognitive focus, achieving a dynamic balance between complexity and comprehensibility at different attention levels.

[0075] Collective Cognitive Enhancement Mechanism: Pioneeringly realizes the integration and enhancement of multi-decision-maker cognitive models. The system can capture and integrate the cognitive patterns of multiple decision-makers, construct a "super-individual cognitive structure" beyond individuals, and can actively identify group cognitive blind spots, guiding decision-makers to pay attention to the neglected decision space areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a flowchart of an intelligent game and operational research optimization assisted decision-making method with enhanced atlas of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0078] Embodiment 1, an intelligent game and operational research optimization assisted decision-making method with enhanced atlas, as Figure 1 shown, includes the following steps:

[0079] Step 100, obtaining multi-source heterogeneous data and inputting it into a spatio-temporal semantic mapping framework to generate a unified spatio-temporal semantic data structure;

[0080] Specifically, multi-source heterogeneous data such as game strategy data, operational research optimization target data, historical decision data, and environmental constraint data are input into the spatio-temporal semantic mapping framework. After processing, a normalized data structure with unified spatio-temporal semantics is output, and this data structure supports unified retrieval and analysis operations across data sources.

[0081] The specific implementation of this step includes:

[0082] Step 101, performing semantic parsing on the input multi-source heterogeneous data to extract semantic features in three dimensions: time, space, and decision;

[0083] Step 102, aligning and fusing data from different sources according to a unified spatio-temporal semantic model, and this model is defined as:

[0084]

[0085] Among them, represents the aligned data, represents the source data, , and respectively represent the semantic descriptors of the time dimension, space dimension, and action dimension. is a semantic alignment function that uses the multi-head self-attention mechanism to achieve the extraction and alignment of cross-domain semantic features.

[0086] Works through the following steps:

[0087] Construct independent attention heads for each dimension (T, S, A);

[0088] Each attention head calculates the semantic similarity under its dimension;

[0089] Merge the attention outputs of each dimension to generate a unified semantic representation.

[0090] Step 103, establish a spatio-temporal semantic index structure to form a spatio-temporal semantic unified database that can be quickly retrieved. This index structure uses an improved R-tree structure, which extends the traditional R-tree to support the hybrid indexing of multi-dimensional spatio-temporal attributes and semantic attributes.

[0091] The unified spatio-temporal semantic data structure output in this step contains the following:

[0092] A set of data items with unified time tags;

[0093] A set of data items with unified space references;

[0094] A set of data items with unified decision semantic tags;

[0095] A spatio-temporal semantic index that supports multi-dimensional retrieval.

[0096] This step is different from traditional data preprocessing that only focuses on the unification of data formats. This method realizes the deep unification at the data semantic level, especially the semantic expression of time, space, and decision-making behaviors, laying a foundation for the subsequent construction of the cognitive-spatio-temporal manifold.

[0097] Step 200, the unified spatio-temporal semantic data structure generates a cognitive-spatio-temporal manifold model through the cognitive spatio-temporal manifold generation algorithm;

[0098] This cognitive-spatio-temporal manifold model is a mathematical representation structure that maps the unified spatio-temporal semantic data into the cognitive space, making the relationships between data conform to human cognitive characteristics and providing a basic representation structure for subsequent human-computer interaction.

[0099] The specific implementation of this step includes:

[0100] Step 201, define the mathematical representation of the cognitive-spatiotemporal manifold. The cognitive-spatiotemporal manifold is defined as:

[0101]

[0102] where represents the cognitive-spatiotemporal manifold, represents the set of vertices on the manifold, corresponding to the key nodes in the data space; represents the set of edges between vertices, indicating the relationships between nodes; represents the mapping matrix from physical spacetime to the cognitive space; represents the metric tensor on the manifold, used to describe the cognitive complexity of different regions on the manifold.

[0103] Step 202, extract the non-linear manifold structure from the unified spatiotemporal semantic data. This process is completed using the cognitive spatiotemporal manifold generation algorithm, which combines the principles of diffusion maps and Riemannian geometry, retains the advantages of diffusion maps in capturing discrete relationships, and utilizes the powerful ability of Riemannian geometry in expressing continuous non-linear spaces to form a unified framework capable of adaptively expressing complex cognitive-spatiotemporal relationships, specifically defined as:

[0104]

[0105] where represents the cognitive-spatiotemporal manifold, represents the manifold generation function, represents the unified spatiotemporal semantic data, represents the smoothing parameter, represents the neighborhood size. This function is implemented by iteratively optimizing the following objective function:

[0106]

[0107] where represents the data point and the similarity between, represents the regularization term of the manifold, is the balance parameter, and are the indices of the data points, where the summation symbol indicates summation over all data point pairs is performed, , is the total number of data points, represents the norm of the vector, and respectively represent data points and mapping coordinates.

[0108] Step 203: Embed the manifold into the visualization space and optimize the manifold according to human cognitive characteristics, including reducing cognitive distortion, enhancing the visual salience of key decision points, balancing the visibility of local details and global structure, etc.

[0109] The specific implementation steps of the cognitive spatio-temporal manifold generation algorithm include:

[0110] Standardize the unified spatio-temporal semantic data to eliminate the dimensional differences of different dimensions, using the classic mean-standard deviation standardization method;

[0111] Construct an initial adjacency graph, and use the k-nearest neighbor algorithm to connect k nearest neighbor points for each data point;

[0112] Calculate the weight matrix of the initial adjacency graph. The weights between neighboring points are based on the Gaussian kernel function of the inter-point distance, and the weights of non-neighboring points are zero;

[0113] Calculate the similarity matrix: Calculate the similarity based on the distance between data points using the Gaussian kernel function;

[0114] Construct the Laplacian matrix, that is, the degree matrix minus the weight matrix:

[0115]

[0116] where is a diagonal matrix, and the diagonal element represents the degree of node i, is the Laplacian matrix.

[0117] Solve the generalized eigenvalue problem to obtain the eigenvector corresponding to the smallest non-zero eigenvalue. The calculation formula is as follows:

[0118]

[0119] where is the mapping matrix, containing the coordinates mapped to the cognitive space, is the eigenvalue.

[0120] Use these eigenvectors as the initial embedding of the manifold;

[0121] Introduce the cognitive preference tensor , and adjust the manifold metric to adapt to human cognitive characteristics:

[0122]

[0123] where is the metric tensor on the manifold, is the local coordinate transformation matrix, is the cognitive preference tensor, which characterizes the cognitive importance in different directions.

[0124] Cognitive preference tensor is constructed in the following way:

[0125]

[0126] where represents the cognitive dimension priority basis tensor, represents the corresponding weight coefficient, represents the number of cognitive dimensions. The cognitive dimension priority basis tensor is obtained from cognitive psychology experimental data, which characterizes the cognitive preference differences of humans in different types of information processing.

[0127] Based on the cognitive metric tensor, the geodesic distance on the manifold is calculated as the distance metric in the cognitive space:

[0128]

[0129] where represents connecting and all possible paths, , represents the tangent vector of the path, represents the minimum value, represents and the distance in the cognitive space.

[0130] The cognitive-spatiotemporal manifold model output in this step includes the following:

[0131] Set of data nodes: representing the key data points in the decision space;

[0132] Set of relationship edges: representing the semantic associations between nodes;

[0133] Cognitive metric tensor: describing the cognitive difficulty in different regions of the manifold;

[0134] Manifold mapping function: defining the mapping from the original data space to the cognitive space.

[0135] The core framework of the cognitive-spatiotemporal manifold model consists of a graph structure representation system, a Riemannian manifold representation system, a graph-manifold mapping framework, a cognitive metric field system, and a spatiotemporal evolution mechanism, specifically including:

[0136] Graph structure representation system:

[0137] Adopts standard graph theory components to achieve discrete relationship representation:

[0138] Vertex set representation: Use heterogeneous graph nodes to represent multiple types of entities. Each vertex is in tensor form, containing a feature vector and attribute metadata.

[0139] Edge set representation: Adopt a multi-edge structure to express complex relationships. Each edge has a weight matrix and a relationship type tensor.

[0140] Graph topological properties: Utilize homology groups, persistent homology, and graph spectra to analyze structural features, and apply community detection to identify functional submodules.

[0141] Riemannian manifold representation system:

[0142] Implement continuous space representation based on standard differential geometry components:

[0143] Manifold structure definition: Use graph embedding and isometric embedding to establish a d-dimensional Riemannian manifold, and use coordinate atlases and transition functions to define local structures.

[0144] Metric tensor field: Define the Riemannian metric tensor to characterize the distance measure on the manifold, and use a positive semi-definite matrix to ensure the validity of the metric.

[0145] Geodesic equation: Derive the geodesic differential equation based on the Euler-Lagrange equation, and apply variational methods to solve the shortest path.

[0146] Curvature properties: Utilize the Riemannian curvature tensor, sectional curvature, and Riemannian scalar to analyze the bending properties of the manifold.

[0147] Graph-manifold mapping framework:

[0148] Combine standard methods of topological data analysis and manifold learning:

[0149] Embedding mapping: Use manifold learning algorithms such as geodesic multi-dimensional scaling, Laplacian eigenmaps, and diffusion maps to establish mapping relationships.

[0150] Structure preservation principle: Follow topological invariance principles such as homeomorphic mapping, isometric mapping, and conformal mapping to preserve the original structural properties.

[0151] Regularization strategy: Adopt entropy regularization, differential regularization, and topological regularization to control the mapping quality.

[0152] Cognitive metric field system:

[0153] Based on standard theories of cognitive science and information geometry:

[0154] Metric field representation: Utilize information metrics, Fisher information matrices, and relative entropy to construct cognitive distance metrics.

[0155] Attention modulation mechanism: Adjust local metric properties based on cognitive control theory and the principle of selective attention.

[0156] Multi-scale representation: Apply wavelet transform and scale space theory to achieve a multi-resolution representation structure.

[0157] Space-time evolution mechanism:

[0158] Integrate dynamic systems and differential equation theory:

[0159] Vector field representation: Use the vector field on the manifold to represent the direction of time evolution, and adopt the Lie derivative to describe the rate of change on the manifold.

[0160] Manifold deformation: Calculate space-time deformation based on flow models, parallel transport, and exponential mapping.

[0161] Evolution equation: Apply partial differential equations, level set methods, and variational integrals to describe the dynamic characteristics of the system.

[0162] This step is different from traditional low-dimensional embedding methods that only focus on maintaining geometric structures. This algorithm also takes into account the laws of human cognition. By means of the metric tensor define the concept of "distance" in the cognitive space, so that in the cognitive space, data points with strong decision-making relevance are visually closer, while data points with weak decision-making relevance are visually more distant, thus creating a decision space representation that conforms to human intuition.

[0163] Step 300, Obtain user interaction data and the cognitive-spacetime manifold model, and generate an adaptive view expression through the interactive response manifold deformation algorithm;

[0164] The adaptive view expression refers to a manifold structure that is dynamically adjusted according to the user's cognitive focus, which can reflect the user's interactive intentions and cognitive needs in real time, enabling a closed-loop interaction between data expression and the user's cognitive process.

[0165] The specific implementation of this step includes:

[0166] Step 301, Obtain the user's interaction operations and convert them into manifold deformation operators:

[0167]

[0168] Among them, represents the user's attention distribution at time t, represents the decision-making operation executed by the user, represents the cognitive-spacetime manifold, and are the attention response operator and the decision response operator respectively, is the manifold deformation operator.

[0169] Step 302, the interactive response manifold deformation algorithm calculates the deformation process of the manifold according to the user's interactive operations. The specific calculation formula is as follows:

[0170]

[0171] This partial differential equation is based on the application of partial differential equations in physical simulations (such as the heat diffusion equation, elastic deformation equation) and the implicit Euler method, which is a standard technique for solving ordinary differential equations in numerical analysis;

[0172] Applying physical simulation technology to the field of cognitive interaction, two new operators are defined:

[0173] : The attention response operator, which captures the influence of the user's attention on the manifold;

[0174] : The decision response operator, which simulates the influence of the user's decision-making behavior on the manifold;

[0175] The specific forms of these two operators are obtained by fitting a large amount of user interaction experimental data.

[0176] To ensure the stability and continuity of the deformation process, an improved implicit Euler method is used for numerical solution:

[0177]

[0178] Among them, represents the manifold state at time t, represents the manifold state at time , and

[0179] The improved implicit Euler method is more suitable for the characteristics of the cognitive-spatiotemporal manifold by specifically processing the attention response operator and the decision response operator compared with the traditional implicit Euler method.

[0180] Step 303, perform corresponding manifold deformation operations on different types of user interaction operations:

[0181] Attention focusing operation: According to the user's attention area, adjust the display ratio and detail level of the corresponding part of the manifold, so that this area expands in the visual space, the details are enhanced, and the context information of the surrounding area is maintained at the same time;

[0182] Decision operation: When it is detected that the user performs a decision operation, calculate the predicted distribution of the decision result, and generate a decision trajectory and a possible result distribution map on the manifold;

[0183] Hypothetical scenario generation: When the user proposes a hypothetical scenario, a corresponding scenario branch structure is created on the manifold, and the decision evolution paths under different hypotheses are calculated.

[0184] The adaptive view expression output in this step includes the following:

[0185] The deformed manifold structure: A manifold model reflecting the user's cognitive focus;

[0186] Attention-enhanced region: A high-detail representation of the user's attention area in the manifold;

[0187] Decision trajectory: The predicted decision result path;

[0188] Scenario branch: Manifold variants under different hypotheses.

[0189] In the traditional human-computer interaction process, data expression and user interaction are two separate processes. However, in this step, the two are organically integrated through the manifold deformation mechanism, enabling the data expression to be dynamically adjusted according to the user's cognitive state and interaction intention, achieving a true "human-computer dialogue" effect.

[0190] Step 400: Obtain the user interaction behavior data and the adaptive view expression, and generate a cognitive balance view through the cognitive load perception projection algorithm;

[0191] The cognitive balance view is a data presentation method that balances complexity and comprehensibility, dynamically adjusting the dimension and detail level of data presentation according to the user's current cognitive load state to ensure the efficiency of information transmission.

[0192] The specific implementation of this step includes:

[0193] Step 401: Analyze the user interaction behavior data and evaluate the current cognitive load level:

[0194]

[0195] Among them, represents the calculated cognitive load level value, represents the user's historical interaction records, represents the user's gaze pattern data, represents the user's operation response time data, is the cognitive load evaluation function. The cognitive load evaluation function adopts a weighted fusion method, and the specific weight coefficients are learned from a large amount of user experiment data through machine learning methods.

[0196] Step 402: Determine the dimension and method of data projection according to the cognitive load evaluation result:

[0197]

[0198] Among them, represents the adaptive projection result, represents the original high-dimensional data, represents the cognitive load level value, represents the current decision context information, is the adaptive projection function.

[0199] The core idea of the cognitive load-aware projection algorithm is to retain more dimensions and details when the cognitive load is low, and reduce dimensions and simplify the expression when the cognitive load is high, while ensuring that key decision information is not lost. The algorithm automatically selects an appropriate dimensionality reduction method according to the cognitive load value, including:

[0200] When adopts the non-linear dimensionality reduction method that retains the most information (such as t-SNE (t-Distributed Stochastic Neighbor Embedding), UMAP (Uniform Manifold Approximation and Projection), Isomap (Isometric Mapping), etc.);

[0201] When adopts the dimensionality reduction method that balances information retention and computational efficiency (such as MDS (Multidimensional Scaling), Sammon Mapping, etc.);

[0202] When adopts the simplest linear dimensionality reduction method (such as PCA (Principal Component Analysis), LDA (Linear Discriminant Analysis), etc.);

[0203] Among them represents the low threshold of cognitive load, represents the high threshold of cognitive load, represents the cognitive load value.

[0204] Step 403, generate a semantic interpretation for the projection result to improve interpretability:

[0205]

[0206] Among them, represents the generated projection interpretation text, represents the adaptive projection result, represents the original data, represents the current context information, is the interpretation generation function.

[0207] The function realizes the conversion from mathematical representation to semantic description through the following four-step process:

[0208] Feature attribution analysis: Identify which original features have the greatest impact on the projection result;

[0209] This step performs dimensional importance quantification, calculating the statistical correlation between the dimensionality reduction result and the original high-dimensional features through a backward feature tracing algorithm. Typical implementations usually involve principal component contribution, global sensitivity analysis, or SHAP value calculation to determine which original features have a dominant influence on the current projection representation. This process conforms to the principle of feature importance attribution in interpretable machine learning.

[0210] Topological structure analysis: Detect the intrinsic topological structure in the projection space;

[0211] This step detects the intrinsic topological structure in the projection space through unsupervised learning methods, dividing the observed data into cohesive subpopulations. Common implementations include density-based spatial clustering, hierarchical clustering, or spectral clustering techniques, and the optimal clustering parameter configuration is determined through internal evaluation metrics such as the silhouette coefficient and Davies-Bouldin index to achieve the adaptive recognition of the topological structure.

[0212] Semantic pattern construction: Map statistical features to domain semantic expressions;

[0213] This step realizes the mapping transformation from statistical features to domain semantics. Through a feature-semantic dictionary and natural language generation templates, the identified data patterns are transformed into semantic expressions that conform to domain terms. This process follows the content planning and surface realization principles in natural language generation (NLG), mapping quantitative statistical features to qualitative semantic descriptions and establishing a bridge between data and domain knowledge.

[0214] Adaptive adjustment of cognitive load: Dynamically adjust the complexity of the explanation according to the user's cognitive state

[0215] In the final stage, the granularity and complexity of the semantic explanation are dynamically adjusted according to the user's cognitive state and decision context. By applying cognitive load theory and relevant principles of information presentation, the system can establish a balance between information completeness and cognitive economy, providing explanatory content with the optimal information density for users with different cognitive needs and professional levels.

[0216] The cognitive balance view output by this step includes the following:

[0217] Adaptive dimensional projection data: Dynamically adjust the data representation of dimensions according to cognitive load;

[0218] Highlighting of key decision features: Highlight the features crucial for decision-making;

[0219] Projection explanation text: Semantic explanation of the current projection view;

[0220] Cognitive load indicator: A quantitative representation of the current user's cognitive state.

[0221] This step is different from traditional methods that improve comprehensibility by simplifying data. This method retains all the complexity of the data, but dynamically adjusts the dimensions and details of data presentation according to the user's cognitive state, achieving a dynamic balance between complexity and comprehensibility. The system does not preset a fixed data representation form, but dynamically generates the most suitable expression for the current situation according to the user's cognitive needs and abilities.

[0222] Step 500: Obtain the interaction data and cognitive balance views of multiple decision-makers, and output the collective cognitive structure and cognitive guidance strategy through the multi-agent cognitive fusion algorithm;

[0223] The collective cognitive structure is a fused representation of the cognitive models of multiple decision-makers, which can capture the complementarity and blind spots of group thinking, while the cognitive guidance strategy is an attention guidance scheme generated for the identified cognitive blind spots.

[0224] The specific implementation of this step is as follows:

[0225] Step 501: Extract the individual cognitive models from the interaction data of each decision-maker:

[0226]

[0227] Among them, represents the constructed i-th user cognitive model, represents the interaction history data of the i-th user, represents the attention distribution data of the i-th user, represents the decision-making behavior data of the i-th user, is the cognitive model extraction function. The cognitive model extraction function is a well-known technology based on user modeling and behavior analysis. The innovation lies in comprehensively analyzing three types of data: interaction history, attention distribution, and decision-making behavior, to construct a complete user cognitive model.

[0228] Step 502: Fuse multiple individual cognitive models into a collective cognitive model:

[0229]

[0230] Among them, represents the generated collective cognitive model, represents the set of n extracted individual cognitive models, represents the fusion weight parameter, represents the current collaboration context information, represents the model fusion function, represents the starting index of the set, Let \(N\) denote the total number of decision-makers. The model fusion function uses graph neural network (GNN) technology, which regards individual cognitive models as nodes in a graph and defines the complementary or conflicting relationships between nodes through a relational adjacency matrix to achieve non-linear cognitive model fusion.

[0231] The multi-agent cognitive fusion algorithm does not perform simple model averaging, but instead achieves weighted non-linear fusion through graph neural networks, capable of capturing the structural relationships between different cognitive models:

[0232]

[0233] where \(GNN(\cdot)\) represents the graph neural network function, \(A\) represents the relational adjacency matrix between models, characterizing the complementary or conflicting relationships between different individual cognitive models, \(I\) represents the set of individual cognitive models from index \(i\) to \(N\), \(C\) represents the collective cognitive model.

[0234] Step 503: Based on the collective cognitive model, identify potential blind spots in group cognition:

[0235]

[0236] where \(B\) represents the set of identified cognitive blind spots, \(C\) represents the collective cognitive model, \(D\) represents the data representation of the complete decision space, \(\theta\) is the detection threshold parameter, \(f_{bd}(\cdot)\) represents the blind spot detection function. A cognitive blind spot refers to an area that exists in the decision space but is weakly represented in the collective cognitive model.

[0237] The blind spot detection function is essentially a combined application of coverage analysis and anomaly detection, and specifically performs the following steps:

[0238] Divide the decision space into grid cells;

[0239] Calculate the "attention density" of the collective cognitive model for each cell;

[0240] Identify areas where the attention density is lower than the threshold but the decision importance is high;

[0241] Mark these areas as the set of cognitive blind spots .

[0242] Step 504: Generate a guidance strategy for the identified cognitive blind spots:

[0243]

[0244] Among them, represents the generated set of cognitive guidance strategies, represents the set of cognitive blind spots, represents the collective cognitive model, represents the current decision context, represents the strategy generation function.

[0245] The strategy generation function specifically performs the following steps:

[0246] Select a suitable guidance template from the strategy library according to the current context and blind spot characteristics;

[0247] Calculate the "attention guidance priority" for each blind spot;

[0248] Generate a guidance strategy including visual cues, interaction suggestions, and explanatory text based on the priority;

[0249] Combine multiple local strategies into a set of cognitive guidance strategies .

[0250] Common strategies include: visual highlighting, dynamic zooming, interaction prompts, comparison frameworks, and semantic explanations, etc.

[0251] The output of this step includes the following:

[0252] Collective cognitive model: A fused representation of the cognitive patterns of multiple decision-makers;

[0253] Cognitive blind spot map: Identifying the key areas in the decision space that are collectively overlooked;

[0254] Cognitive guidance strategy: An attention guidance scheme generated for cognitive blind spots;

[0255] Cognitive complementarity index: Quantifying the degree of complementarity between the cognitive patterns of different decision-makers.

[0256] This step is different from traditional collaborative decision-making methods that only focus on the summary of opinions and the resolution of differences. This method deeply analyzes and integrates the cognitive processes of decision-makers. It can not only integrate the cognitive advantages of multiple decision-makers, but also actively identify the blind spots in collective thinking, and guide the attention of decision-makers to these overlooked areas through guidance strategies, thus realizing the excavation and enhancement of collective wisdom.

[0257] Step 600: Generate an interactive decision-making scenario and a multi-modal interaction interface through immersive spatio-temporal scene rendering technology.

[0258] This step takes the outputs of the previous steps (cognitive-spatiotemporal manifold model, adaptive view expression, cognitive balance view, collective cognitive structure, and cognitive guidance strategy) as inputs, and through immersive spatiotemporal scene rendering technology for conversion processing, generates an interactive decision-making scenario and a multimodal interaction interface.

[0259] An interactive decision-making scenario refers to a visual environment that can intuitively display the decision-making space, support user operations and simulations, while the multimodal interaction interface provides multiple human-computer interaction methods, enabling decision-makers to interact with the decision-making scenario naturally.

[0260] The specific implementation of this step is as follows:

[0261] Step 601, perform spatiotemporal scene mapping to convert the cognitive-spatiotemporal manifold into a visual space representation:

[0262]

[0263] Among them, represents visual scene data, represents the cognitive-spatiotemporal manifold model, represents the visual style parameter set, represents the user preference configuration, is the scene mapping function.

[0264] Step 602, based on the visual space representation, perform interactive scene rendering:

[0265]

[0266] Among them, represents the generated interactive scene, represents the visual scene data, represents the interaction parameter configuration, represents the physical dynamics parameter set, is the scene rendering function.

[0267] The scene mapping function and the scene rendering function are based on well-known technologies in the fields of computer graphics and visualization, including: information visualization mapping technology, physical engines and dynamics simulation, and interactive rendering technology.

[0268] Step 603, construct a multimodal interaction interface to implement the following interaction methods:

[0269] (1) Gesture interaction module: Capture user gestures and convert them into scene operation instructions;

[0270] (2) Voice interaction module: Parse user voice instructions and execute corresponding scene control and query functions;

[0271] (3) Collaborative Interaction Module: Synchronize multi-user operations and maintain a consistent scenario state.

[0272] The input of each interaction method is converted into corresponding manifold deformation operations through a unified interaction-response framework, realizing the seamless connection between user intentions and data expressions.

[0273] Step 604, Provide decision path planning and simulation functions:

[0274]

[0275] Among them, represents the generated decision path data, represents the action sequence specified by the user, represents the cognitive-spatiotemporal manifold model, represents the set of applied constraint conditions, represents the decision simulation function.

[0276] The decision simulation function integrates Monte Carlo simulation and Markov decision process (MDP), and the specific execution process is as follows:

[0277] Map the action sequence specified by the user to the trajectory on the manifold;

[0278] Consider the actionable actions under the constraint conditions

[0279] at each decision point;

[0280] Use a hybrid model (combining deterministic and probabilistic models) to predict decision results:

[0281] For the results under deterministic constraints, use a causal model to calculate;

[0282] For uncertain factors, use Monte Carlo simulation to generate the probability distribution of possible results; Generate decision path data

[0283] including:

[0284] The main path (the most likely result);

[0285] The uncertainty boundary (the probability distribution of possible results);

[0286] Key decision points (nodes with high path sensitivity).

[0287] These paths are visualized as time-evolving trajectories on the manifold, enabling decision-makers to intuitively evaluate the long-term impact of decision sequences.

[0288] Interactive decision-making scenario: an operable three-dimensional or multi-dimensional scenario environment;

[0289] Multi-modal interaction interface: a human-computer interaction interface that supports gestures, voice, and collaborative operations;

[0290] Decision simulation engine: a tool component used for calculating and visualizing decision results;

[0291] Multi-user collaboration layer: a collaborative infrastructure that supports the simultaneous participation of multiple decision-makers.

[0292] The interactive decision-making scenario in this step has the following characteristics:

[0293] Dynamic responsiveness: The scenario can respond to user operations in real time and provide instant feedback;

[0294] Multi-scale navigation: Supports seamless switching from macroscopic overviews to microscopic details;

[0295] Predictive display: Actively predicts and visualizes the possible results of user operations;

[0296] Collaborative consistency: Maintains the consistency of multi-user views and operation synchronization.

[0297] This step is different from traditional data visualization technologies that only provide static or limited interaction views. This method transforms the abstract decision space into an intuitive spatio-temporal scenario and supports rich interaction methods, enabling decision-makers to intuitively perceive, manipulate, and plan decisions as if in the physical world, truly realizing the "immersive decision-making" experience.

[0298] Technical effects of this embodiment:

[0299] By integrating abstract spatio-temporal data analysis with the human cognitive process into a unified interactive cognitive-spatio-temporal manifold, this embodiment realizes a paradigm shift from "passive data display" to "active cognitive guidance", resulting in the following significant technical effects:

[0300] Improved decision understanding efficiency: The system transforms complex game and operations research optimization problems into intuitive spatio-temporal scenarios, enabling decision-makers to quickly understand the essence of the problems. Practical applications show that the understanding time for complex game decisions using this system is reduced by 81%, that is, decision-makers can master decision problems of the same complexity in less than one-fifth of the time.

[0301] Significantly reduced cognitive load: Through cognitive adaptive projection technology, the system can dynamically adjust the data presentation method according to the user's cognitive state, effectively reducing the cognitive load. Tests show that when processing 64-dimensional high-dimensional data, the user's cognitive load score is reduced by 68%, while maintaining 92% data information fidelity.

[0302] Improved Team Collaboration Efficiency: The collective cognition enhancement mechanism enables team members to share cognitive patterns and complement each other's cognitive blind spots, resulting in a 156% increase in team collaboration efficiency. This increase is even more significant, reaching 203%, especially in heterogeneous teams (members with different professional backgrounds).

[0303] Significantly Improved Decision-Making Quality: By visualizing decision paths and result distributions, decision-makers can more comprehensively evaluate the impact of decisions, leading to a 43% improvement in decision-making quality. This improvement is particularly notable in decision-making scenarios that require balancing short-term and long-term benefits.

[0304] Effectively Eliminated Decision Blind Spots: The system can actively identify and guide decision-makers to pay attention to overlooked decision factors. Experiments show that the system can detect 94% of the decision blind spots in traditional methods, effectively preventing decision biases and risk neglect.

[0305] Breakthrough in High-Dimensional Complex Decision-Making Capability: The system has successfully supported the visualization and interaction of complex decision-making problems involving 64 dimensions and 12 time scales, breaking through the dimensional limitations of traditional visualization methods and enabling decision-makers to intuitively grasp the ultra-high-dimensional decision space.

[0306] Immersive Decision-Making Experience: The interactive spatio-temporal scenarios provided by the system make the decision-making process intuitive and immersive. The user satisfaction score has increased by 187% compared to traditional decision support systems, and the willingness to use the system has increased by 212%.

[0307] Accelerated Knowledge Transfer: By visualizing abstract decision-making problems, the system has significantly accelerated the knowledge transfer between domain experts and novices. The understanding speed of novices for complex decision-making problems has increased by 276%, and their decision-making ability has increased by 152%.

[0308] The core breakthrough of this implementation lies in establishing a dynamic balance mechanism between data complexity and cognitive comprehensibility, making complex decision-making problems intuitive and controllable without sacrificing the complexity and integrity of the problems. Instead of simplifying data to fit cognition, the system dynamically adjusts the data presentation dimensions and details according to cognitive foci through cognitive adaptive projection technology, enabling complexity and comprehensibility to coexist at different attention levels and fundamentally solving the problem of understanding and decision-making for high-dimensional complex data.

[0309] An application example of Embodiment 1 is as follows:

[0310] Urban traffic management departments face complex challenges in optimizing traffic flow decisions, requiring collaborative decision-making across multiple time scales (real-time, short-term, and long-term) based on multi-source data, including real-time traffic volume, historical traffic patterns, weather conditions, special event arrangements, etc. Decision-makers include traffic control center operators, urban planners, and emergency response teams, who need to quickly identify patterns, predict trends, and develop intervention measures in complex high-dimensional traffic data.

[0311] Traditional traffic management systems usually only provide basic data visualization, lacking intuitive display of high-dimensional data and making it difficult to support collaborative analysis and decision-making among multiple decision-makers. At the same time, a large amount of traffic data often leads to information overload, making it difficult for decision-makers to quickly identify key information and respond.

[0312] This system achieves a dynamic balance between data complexity and cognitive comprehensibility by mapping complex traffic data onto a cognitive-spatiotemporal manifold and dynamically adjusting the data presentation according to the user's cognitive state, significantly improving the efficiency and quality of traffic decision-making.

[0313] Example of input data:

[0314] The multi-source heterogeneous data processed by the system includes:

[0315] Some examples of real-time traffic flow data are shown in Table 1:

[0316] Table 1: Real-time traffic flow data (partial examples)

[0317]

[0318] Some examples of historical traffic pattern data are shown in Table 2:

[0319] Table 2: Historical traffic pattern data (partial examples)

[0320]

[0321] Some examples of environment and event-related data are shown in Table 3:

[0322] Table 3: Environment and event data (partial examples)

[0323]

[0324] Some examples of the traffic intervention measure database are shown in Table 4:

[0325] Table 4: Intervention measure database (partial examples)

[0326]

[0327] Verification of technical effects:

[0328] To verify the technical effects of this embodiment in the urban traffic flow management scenario, we compared the decision-making efficiency and quality performance of the traditional traffic management platform and this system in three representative traffic management scenarios. The verification process adopted a controlled experiment method, and 20 traffic management professionals used the traditional platform and this system respectively to complete the same decision-making tasks.

[0329] Verification of the improvement in decision-making understanding efficiency:

[0330] Verification object: Whether this system can significantly improve the decision-makers' understanding efficiency of complex traffic data.

[0331] The experimental results of the comparison of decision-making understanding efficiency are shown in Table 5:

[0332] Table 5: Comparison results of decision-making understanding efficiency

[0333]

[0334] It can be seen from the test results that compared with the traditional system, this system has significantly improved the decision-makers' understanding efficiency of complex traffic situations:

[0335] In the peak-hour traffic hot spot identification task, the number of hot spots that the decision-makers can identify has increased by 126.9%, and the time required for identification has decreased by 73.3%;

[0336] In the analysis of the impact of special events, the accuracy rate of the decision-makers' estimation of the impact scope has increased by 32.1%, and the recognition rate of the secondary impact area (such as the congestion diffusion area) has increased by 89.3%;

[0337] In the abnormal traffic pattern recognition, the abnormal recognition accuracy rate of the decision-makers has increased by 37.4%, and the time for proposing response measures has decreased by 72.4%.

[0338] These results verify that through the cognitive-spatiotemporal manifold model and the interaction-induced manifold deformation mechanism, this system effectively reduces the cognitive burden of complex data understanding, enabling decision-makers to quickly and accurately grasp the overall traffic situation, identify key problem points and their causes.

[0339] Verification of the improvement in multi-agent collaborative decision-making efficiency:

[0340] Verification object: Whether this system can significantly improve the efficiency of multi-decision-makers in collaborative solving of complex traffic problems.

[0341] The comparison results of multi-agent collaborative decision-making efficiency are shown in Table 6:

[0342] Table 6: Comparison results of multi-agent collaborative decision-making efficiency

[0343]

[0344] The experimental results show that this system exhibits significant advantages in the scenario of multi-agent collaborative decision-making:

[0345] In the formulation of traffic control plans for large-scale events, the team using this system saved 72.4% of the time compared to the traditional method, and the number of negotiation rounds decreased by 61.4%;

[0346] In the joint response to traffic emergencies, the time for formulating response plans decreased by 76.7%, and the cross-departmental coordination time decreased by 78.2%;

[0347] In the adaptive adjustment of seasonal traffic patterns, the multi-department collaborative planning time decreased by 72.0%, and at the same time, the planning adaptability score increased by 24.1%.

[0348] These results verify that through the collective cognition enhancement mechanism, this system effectively integrates the professional perspectives of different decision-makers, identifies and makes up for the cognitive blind spots in collective decision-making, and significantly improves the efficiency and quality of team collaborative decision-making. In addition, the sharing mechanism of the interactive decision-making scenario enables all decision-makers to collaborate within the same cognitive space, greatly reducing information asymmetry and communication costs.

[0349] Verification of decision quality improvement:

[0350] After this system was actually deployed in the "Intelligent Transportation Management Center of City M", a three-month follow-up evaluation was conducted on the impact of the system on the decision-making quality of urban traffic management. By comparing the changes in key performance indicators of traffic management before and after the system deployment, the effect of decision quality improvement was verified.

[0351] The comparison results of decision-making quality in the three months before and after the system deployment are shown in Table 7:

[0352] Table 7: Comparison results of decision-making quality (average values in the three months before and after the system deployment)

[0353]

[0354] The actual deployment results verify the significant improvement of this system in decision-making quality:

[0355] In terms of traffic congestion management: The average speed of the main roads during peak hours increased by 22.5%, and the cross-regional congestion diffusion events decreased by 63.0%;

[0356] In terms of traffic event response: The average response time to events decreased by 63.0%, and the effectiveness rate of intervention measures increased by 22.1%;

[0357] In terms of resource utilization: The balance degree of road network capacity utilization increased by 43.1%, and the optimization frequency of signal control increased by 195.7%;

[0358] In terms of user experience: The predictability of the travel time reported by drivers has increased by 35.4%, and the public transportation satisfaction has increased by 26.6%.

[0359] These actual operation data indicate that this system not only significantly improves the understanding efficiency of complex traffic data and the multi-agent collaborative decision-making efficiency of decision-makers, but more importantly, these efficiency improvements have been transformed into the improvement of the actual traffic management quality. By providing more intuitive and comprehensive decision-making support, the system enables decision-makers to formulate more accurate and effective traffic management strategies, thus achieving the overall optimization of urban traffic flow.

[0360] The application of this implementation method in the urban traffic flow management scenario verifies the following key technological innovation points and their effects:

[0361] Cognitive-spatiotemporal unified expression of heterogeneous data: Transforming multi-source heterogeneous traffic data into a unified cognitive-spatiotemporal manifold model verifies its significant effect in solving the problems of data fragmentation and multi-source heterogeneous data fusion. In actual applications, the completeness of decision-makers' understanding of traffic conditions has increased by 87.3%.

[0362] Manifold deformation mechanism induced by interaction: Realizing the dynamic adaptation of data expression and the user's cognitive focus verifies the effectiveness of highlighting key decision-making areas while maintaining a global perspective. Experimental results show that the accuracy of data understanding in the areas concerned by decision-makers has increased by 91.5%, and at the same time, the ability to grasp the overall situation has increased by 63.8%.

[0363] Adaptive projection aware of cognitive load: Dynamically adjusting the data presentation dimension and detail level according to the user's current cognitive state verifies its effect in balancing data complexity and comprehensibility. Tests show that the decision-making accuracy of users in a high cognitive load state has increased by 76.2%.

[0364] Collective cognition enhancement and blind spot compensation: Effectively integrating the professional perspectives of multiple decision-makers and identifying and compensating cognitive blind spots verifies its significant effect in improving the efficiency and quality of team collaborative decision-making. In actual applications, the cross-departmental collaboration efficiency has increased by 75.6%, and the recognition rate of collective decision-making blind spots has increased by 82.3%.

[0365] Interaction in immersive spatiotemporal decision-making scenarios: Providing an intuitive and immersive decision-making experience verifies its effect in reducing the cognitive burden of complex decision-making. Tests show that the cognitive burden of decision-makers in complex scenario analysis has been reduced by 68.5%, and the decision-making confidence has increased by 56.7%.

[0366] This embodiment applies the "Co - creative Decision - making Scenario System of Hyper - dimensional Spacetime Atlas" to the urban traffic flow management scenario. Through the comprehensive effect of the above - mentioned technological innovation points, it has successfully achieved the technical goal of "expressing complex game - theory and operational research optimization problems with intuitive spacetime scenarios", made a breakthrough in balancing data complexity and cognitive comprehensibility, and significantly improved the decision - making efficiency and quality in complex decision - making scenarios.

[0367] The above describes the embodiments of the present invention. However, these embodiments are not limited to the above - mentioned specific implementation manners. The above - mentioned specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A graph-enhanced intelligent game and operations optimization decision-making assistance method, characterized in that: The following steps are involved: Obtain multi-source heterogeneous data and input them into the spatiotemporal semantic mapping framework to generate a unified spatiotemporal semantic data structure; The unified spatiotemporal semantic data structure generates a cognitive-spatiotemporal manifold model through a cognitive spatiotemporal manifold generation algorithm; The cognitive space-time manifold generation algorithm includes: Standardize the unified spatiotemporal semantic data to eliminate the dimensional differences of different dimensions, using the classic mean-standard deviation standardization method; Construct an initial adjacency graph and use the k-nearest neighbor algorithm to connect the k nearest neighbors for each data point; Calculate the weight matrix of the initial adjacency graph. The weights between neighboring points are based on the Gaussian kernel function of the distance between points, and the weights of non-neighboring points are zero. Calculate the similarity matrix: Based on the distance between data points, use the Gaussian kernel function to calculate the similarity; Construct the Laplacian matrix, which is the degree matrix minus the weight matrix; Solve the generalized eigenvalue problem and obtain the eigenvector corresponding to the smallest non-zero eigenvalue; Introducing cognitive preference tensor , adjust the manifold metric to accommodate human cognitive characteristics: ； in is the metric tensor on the manifold, is the local coordinate transformation matrix, is the cognitive preference tensor, which characterizes the cognitive importance in different directions; Cognitive Preference Tensor Constructed by: ； in represents the cognitive dimension priority basis tensor, represents the corresponding weight coefficient, Indicates the number of cognitive dimensions; Compute the geodesic distance on the manifold based on the epistemic metric tensor as a distance metric in epistemic space: ； in Indicates connection and All possible paths of , represents the tangent vector of the path, Indicates the minimum value, express and distance in cognitive space; Obtain user interaction data and cognitive-space-time manifold model, and generate adaptive view expression through interactive response manifold deformation algorithm; The specific implementation of this step includes: Step 301, obtain the user interaction operation and convert it into a manifold deformation operator: ； in, represents the user attention distribution at time t, represents the decision operation performed by the user, Represents the cognitive-space-time manifold model, and They are the attention response operator and the decision response operator respectively. is the manifold deformation operator; Step 302, the interactive response manifold deformation algorithm calculates the deformation process of the manifold according to the user's interactive operation; the manifold deformation process is calculated by the following partial differential equation: ； in represents the user attention distribution at time t, represents the decision operation performed by the user, represents the attention response operator, represents the decision response operator, Representation of cognitive-space-time manifold model; The improved implicit Euler method is used for numerical solution: ； in, represents the state of the manifold at time t, express The manifold state at the moment, is the time step; Step 303: perform corresponding manifold deformation operations for different types of user interaction operations: Attention focusing operation: According to the user's attention area, the display ratio and detail level of the corresponding part of the manifold are adjusted so that the area is expanded in the visual space and the details are enhanced, while maintaining the contextual information of the surrounding area; Decision operation: When a user is detected to perform a decision operation, the predicted distribution of the decision results is calculated, and the decision trajectory and possible result distribution map are generated on the manifold; Hypothetical scenario generation: When a user proposes a hypothetical scenario, a corresponding scenario branch structure is created on the manifold and the decision evolution path under different assumptions is calculated Obtain user interaction behavior data and adaptive view expression, and generate cognitive balance view through cognitive load-aware projection algorithm; The cognitive balance view is a data presentation method that strikes a balance between complexity and comprehensibility, dynamically adjusting the dimensions and level of detail of data presentation based on the user's current cognitive load status; The cognitive load perception projection algorithm is implemented by the following steps: analyzing the user interaction behavior data to calculate the cognitive load level value: ； in represents the calculated cognitive load level value, Represents the user's historical interaction records. represents the user's gaze pattern data, Indicates the user's operation response time data. Evaluate functions for cognitive load; Determine the dimension and method of data projection according to the cognitive load level value: ； in represents the original high-dimensional data, Represents the current decision context information, represents the adaptive projection result, is the adaptive projection function; Generate semantic explanations for the projection results: ； in Represents the original data, Represents the generated projection interpretation text, Generate functions for interpretation; Through immersive space-time scene rendering technology, interactive decision-making scenarios and multimodal interaction interfaces are generated.

2. The graph-enhanced intelligent game and operations optimization decision-making auxiliary method according to claim 1 is characterized in that: The spatiotemporal semantic mapping framework adopts a multi-head self-attention mechanism to achieve cross-domain semantic feature extraction and alignment: ； in Indicates the aligned data. Represents the source data, , and Represent the semantic descriptors of time dimension, space dimension and action dimension respectively, is the semantic alignment function.

3. The graph-enhanced intelligent game and operations optimization decision-making auxiliary method according to claim 1 is characterized in that: The cognitive load-aware projection algorithm automatically selects an appropriate dimensionality reduction method based on the cognitive load value: when When , the nonlinear dimensionality reduction method that retains the most information is used; when When , a dimensionality reduction method that balances information retention and computational efficiency is used; when When , the simplest linear dimensionality reduction method is used; in Indicates a low threshold of cognitive load, Indicates a high threshold of cognitive load, Indicates the cognitive load level value.

4. The graph-enhanced intelligent game and operations optimization decision-making auxiliary method according to claim 3 is characterized in that: The immersive spatiotemporal scene rendering technology is implemented by the following steps: performing spatiotemporal scene mapping to convert the cognitive balance view into a visual space representation: ； in Represents a set of visual style parameters, Indicates user preference configuration, is the scene mapping function, represents visual scene data, Representation of cognitive-space-time manifold model; Based on the visual space representation, interactive scene rendering is performed to generate an interactive scene: ； in Indicates the interaction parameter configuration. represents the set of physical dynamic parameters, represents the generated interactive scene, Rendering function for the scene.

5. The graph-enhanced intelligent game and operations optimization decision-making auxiliary method according to claim 4 is characterized in that: After generating the cognitive balance view, the method further includes: Obtaining the interaction data of multiple decision makers and the cognitive balance view, and generating a collective cognitive structure and cognitive guidance strategy through a multi-agent cognitive fusion algorithm; The collective cognitive structure and cognitive guidance strategy are provided as additional input to the immersive spatiotemporal scene rendering technology to enhance the collaborative decision-making ability of the interactive decision-making scenario.

6. The graph-enhanced intelligent game and operational optimization decision-making auxiliary method according to claim 5 is characterized in that: The multi-agent cognitive fusion algorithm is implemented by the following steps: Extract individual cognitive models from the interaction data of each decision maker: ； in, represents the constructed i-th user cognitive model, represents the interaction history data of the i-th user, represents the attention distribution data of the i-th user, represents the decision behavior data of the i-th user, Extract functions for cognitive models; Multiple individual cognitive models are integrated into a collective cognitive model through the graph neural network algorithm: ； in, represents a graph neural network function, Represents the adjacency matrix between models, characterizing the complementary or conflicting relationships between different individual cognitive models. Indicates from index arrive A collection of individual cognitive models, represents the total number of decision makers, represents a collective cognitive model; Identify potential blind spots in group cognition based on the collective cognition model: ； in, represents the set of identified cognitive blind spots, A data representation representing the complete decision space, is the detection threshold parameter, represents the blind spot detection function; Generate guidance strategies for identified cognitive blind spots: ； in, represents the set of cognitive guidance strategies generated, represents the set of cognitive blind spots, represents the current decision context, represents the policy generation function.

7. A graph-enhanced intelligent game and operations optimization decision-making support system, characterized in that: The method is used to execute the graph-enhanced intelligent game and operational optimization decision-making assistance method described in any one of claims 1 to 6, comprising: The spatiotemporal semantic mapping module is used to convert multi-source heterogeneous data into a unified spatiotemporal semantic data structure; A cognitive manifold generation module, which is used to transform the unified spatiotemporal semantic data structure into a cognitive-spatiotemporal manifold model; The manifold deformation module is used to adjust the cognitive-spatiotemporal manifold model according to user interaction data to generate adaptive view representations; The cognitive load perception module is used to analyze user interaction behavior data, dynamically adjust the dimensions and details of data projection, and generate a cognitive balance view; A spatiotemporal scene rendering module for converting cognitive balance views into interactive decision-making scenarios; Multimodal interaction module, used to provide gesture interaction, voice interaction and collaborative interaction functions; The decision simulation module is used to support users in planning decision paths and performing simulations in scenarios.

Images