Intelligent information scheduling operation scene cognitive model construction method

Abstract

The invention relates to the field of information scheduling, and particularly discloses an intelligent information scheduling operation scene cognitive model construction method. Comprising the following steps: multi-modal perception data acquisition, feature fusion and knowledge representation based on neural symbol reasoning, entropy reduction-oriented uncertainty quantification and cognitive depth enhancement, cognitive bidirectional alignment man-machine cooperation framework construction, and intelligent information scheduling strategy generation and dynamic adjustment based on a cognitive model. According to the method, neural symbol reasoning and multi-modal sensing are fused, an entropy reduction oriented uncertainty quantification mechanism is designed, a cognitive bidirectional alignment man-machine cooperation framework is constructed, normal form transformation of a scheduling system from passive response to active cognition is achieved, the accuracy, dynamic adaptability and man-machine cooperation efficiency of intelligent information scheduling are remarkably improved, and the method is suitable for popularization and application. The intelligent information scheduling technology is promoted to advance to a new height, and the increasing complex scheduling requirement is met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information scheduling, and particularly relates to a method for constructing an intelligent information scheduling operation scene cognitive model. BACKGROUND

[0002] In the field of intelligent information scheduling, with the increasing complexity of business and the explosive growth of data scale, the limitations of traditional information scheduling methods are increasingly prominent, mainly in the following aspects:

[0003] Insufficient scheduling accuracy: Traditional scheduling methods are mostly based on fixed rules and simple algorithms, which are difficult to accurately grasp the complex characteristics and dynamic changes of operation scenes, leading to unreasonable resource allocation, low task execution efficiency and other problems. For example, in communication network traffic scheduling, in the face of sudden traffic peaks and changing user demands, traditional methods often fail to identify the scene in time and accurately, resulting in congestion control failure, degradation of service quality, etc.

[0004] Lack of cognitive ability: Although existing intelligent scheduling models introduce machine learning technology, there are still significant defects in the operation scene cognition level. Most models can only perform shallow identification on a single scene, lack the ability to mine and comprehensively analyze deep-level features of the scene, and are difficult to build a comprehensive and accurate scene cognition system, which cannot meet the complex and variable intelligent information scheduling needs.

[0005] Poor dynamic adaptability: Traditional models have limited adaptability to real-time changes and uncertainty factors in operation scenes. They often need human intervention or retraining to cope with new scenes, and cannot dynamically adjust themselves, which is particularly prominent in fields with rapid changes in operation scenes such as industrial production scheduling, often leading to lagging scheduling strategies, affecting production efficiency and resource utilization.

[0006] Low human-machine collaboration efficiency: The collaboration between existing intelligent scheduling systems and human experts lacks effective bidirectional cognitive alignment mechanisms. Machine outputs are difficult for humans to intuitively understand, and human experience is difficult to effectively integrate into machine models, resulting in the inability to fully utilize the advantages of both sides and the difficulty of realizing the advantages of human-machine collaboration.

[0007] Therefore, there is an urgent need for an intelligent information scheduling operation scene cognitive model construction method to solve the above problems. SUMMARY

[0008] The present application aims to solve the problems in the prior art and proposes an intelligent information scheduling operation scene cognitive model construction method, which fuses neural-symbolic reasoning and multi-modal perception, designs an entropy reduction-oriented uncertainty quantification mechanism, constructs a cognitive bidirectional alignment human-machine collaboration framework, realizes the paradigm shift of the scheduling system from "passive response" to "active cognition", and provides a systematic solution for the technological innovation in the field of intelligent scheduling.

[0009] To achieve the above object, the present application adopts the following technical solutions:

[0010] A method for constructing an intelligent information scheduling operation scene cognitive model, comprising the following specific steps:

[0011] S1: Multi-modal perception data acquisition: Synchronously acquire multi-modal perception data with time and space stamps related to intelligent information scheduling from multi-modal perception sources, the time and space stamps being used to ensure alignment and relevance of the data in time and space dimensions and provide a basis for subsequent fusion processing;

[0012] S2: Feature fusion and knowledge representation: Input the multi-modal perception data into a pre-constructed neural-symbolic reasoning network, the neural-symbolic reasoning network being composed of a neural network module and a symbolic reasoning module, respectively used for extracting deep features of each modal data and performing semantic abstraction and knowledge fusion on the extracted deep features by using a domain knowledge graph and logical rules to generate a structured operation scene knowledge representation;

[0013] S3: Uncertainty quantification and cognitive depth enhancement: For the generated operation scene knowledge representation, construct an uncertainty quantification model, calculate the uncertainty measure of each element and its relationship in the operation scene knowledge representation, and then adjust the neural-symbolic reasoning network parameters and the knowledge graph structure through an optimization algorithm guided by entropy reduction, so as to gradually reduce the uncertainty quantification index, thereby enhancing the fine understanding and deep cognitive ability of the cognitive model to the operation scene, enabling the model to more accurately capture potential laws and key details in the scene and realize iterative improvement of cognitive depth;

[0014] S4: Construction of man-machine collaboration framework: Build a man-machine collaboration framework to realize two-way alignment of machine cognitive model and human scheduling expert cognition, visualize the cognitive model on the machine side, collect expert feedback on the human side and convert it into knowledge representation to feed back to the neural-symbolic reasoning network and the uncertainty quantification model, form a closed-loop interaction of man-machine cognition, and construct a complete operation scene cognitive model;

[0015] S5: Intelligent information scheduling strategy generation and dynamic adjustment: Use the constructed cognitive model to realize real-time cognitive prediction of the operation scene, generate scheduling strategy suggestions, monitor the scheduling execution process feedback evaluation, and dynamically adjust the strategy.

[0016] As a further technical solution of the present application, in S1, the perception sources include but are not limited to visual sensors, auditory sensors, environmental sensors, and business system interaction log recording devices; the perception data cover multi-dimensional contents such as object state information, environmental state information, and scheduling operation interaction information in the operation scene.

[0017] As a further technical solution of the present application, in the S2, the knowledge representation contains key elements such as scene objects and their relationships, scheduling operation constraint conditions, and scene dynamic evolution rules, and realizes the conversion from perception data to inferable knowledge, thereby providing a semantic-rich knowledge base for the cognitive model.

[0018] As a further technical solution of the present application, in the S2, the neural-symbolic reasoning network adopts a hierarchical fusion strategy in the feature fusion and knowledge representation process, first performs feature extraction and preliminary fusion on data within the same modality to generate same-modality feature representation, then performs deep fusion on different-modality feature representations through cross-modality feature association and semantic mapping mechanism, and finally forms comprehensive operation scene knowledge representation, thereby ensuring effective integration and complementary use of different modality information.

[0019] As a further technical solution of the present application, in the S3, the uncertainty measurement includes multi-dimensional quantitative indicators such as probabilistic uncertainty, fuzziness, and inconsistency; and the adjustment process of the optimization algorithm comprehensively considers data-driven statistical characteristic constraints and domain knowledge-guided logical constraints.

[0020] As a further technical solution of the present application, in the S3, the uncertainty quantification model, when calculating the uncertainty measurement, combines the dynamic characteristics of the scene, adopts a time sequence sliding window mechanism, and performs uncertainty analysis and tracking on scene knowledge representation at different times, so as to more accurately capture the dynamic uncertainty change rule of the scene, thereby providing a more accurate quantitative basis for the entropy reduction-oriented optimization process and enhancing the cognitive adaptation ability of the cognitive model to the dynamic evolution of the scene.

[0021] As a further technical solution of the present application, in the S4, in the human-machine collaboration framework, at the machine end, the internal state, reasoning process, and decision basis of the cognitive model are visualized and transformed into explainability, so that human experts can intuitively understand the cognitive logic of the machine; at the human end, an interactive feedback mechanism is designed to collect feedback information such as cognitive judgment of human experts on the operation scene, experience knowledge, and correction opinions on machine decisions, and transform these feedback information into formalized knowledge representation, thereby forming a closed-loop interaction of human-machine cognition, continuously improving the human-machine collaboration efficiency through continuous bidirectional collaboration and knowledge sharing, fully exerting the experience advantages of human experts and the cognitive advantages of machines, and jointly optimizing the performance and adaptability of the intelligent information scheduling operation scene cognitive model.

[0022] As a further technical solution of the present application, in the S4, a human-machine collaboration framework is designed, and a human-machine collaborative optimization mechanism based on reinforcement learning is designed, according to the reward signal fed back by a human expert and the performance index of a machine cognitive model, the human-machine collaboration mode and the knowledge fusion strategy are dynamically adjusted, the human-machine collaboration process is encouraged to learn from each other and complement each other's advantages, the overall performance and generalization ability of the intelligent information scheduling operation scene cognitive model are continuously improved, and the human-machine collaborative efficiency is continuously optimized and improved.

[0023] As a further technical solution of the present application, the S5 specifically comprises: using the constructed cognitive model to perform real-time cognition and prediction analysis on the intelligent information scheduling operation scene, generating a corresponding scheduling strategy suggestion according to the cognition result, and dynamically adjusting the scheduling strategy based on the continuous monitoring and feedback evaluation of the scheduling execution process by the cognitive model, so that the intelligent information scheduling system is changed from a "passive response" to an "active cognition" paradigm, the scientificity, accuracy and real-time performance of the scheduling decision are improved, and the scheduling system is adapted to the complex and changeable operation scene requirements.

[0024] As a further technical solution of the present application, the model performance evaluation and verification step is further included, a multi-dimensional evaluation index system is adopted, including but not limited to cognition accuracy, scheduling decision effectiveness, uncertainty reduction degree, and human-machine collaboration synergy, and real operation scene data and simulation test environment are combined to comprehensively evaluate and verify the constructed cognitive model, so that the reliability and effectiveness of the cognitive model in actual application are ensured, the model is iteratively optimized and improved according to the evaluation result, and the complex application requirements in the field of intelligent information scheduling are met.

[0025] The present application has the following advantages:

[0026] 1. Significantly improve scheduling accuracy: through multi-modal perception data acquisition and deep fusion, combined with neural symbolic reasoning, all kinds of information in the operation scene can be perceived in all directions and fine granularity, and the scene characteristics can be accurately grasped; the scheduling strategy generated on this basis fully considers the scene details and complex factors, realizes the more optimal matching of resources and tasks, effectively improves the scheduling accuracy, and reduces resource waste and task delay.

[0027] 2. Enhance model cognitive ability: fuse the feature extraction and knowledge representation methods of neural network and symbolic reasoning, deeply mine the semantic information and logical relationship in scene data, and construct a structured knowledge graph; the uncertainty quantification and cognitive optimization mechanism guided by entropy reduction further improves the depth of the model's cognition of the scene, so that the model can more accurately understand the essence of the scene, and provide a more solid cognitive foundation for intelligent scheduling decision.

[0028] 3. Improve dynamic adaptability: introduce timing sliding window and reinforcement learning mechanism, the model monitors the dynamic changes of the scene in real time, quickly captures uncertain factors; the strategy generation and dynamic adjustment module based on the cognitive model can autonomously optimize the scheduling strategy according to real-time feedback, ensure that the system can quickly respond appropriately when facing scene mutations, maintain efficient scheduling ability, and enhance the flexibility and robustness of intelligent information scheduling.

[0029] 4. Improve human-computer cooperation efficiency: build a cognitive two-way alignment human-computer cooperation framework to realize seamless interaction between machine cognition and human expert experience; the machine visualizes the internal state for human understanding, while collecting human feedback and integrating it into model optimization to form an efficient closed-loop cooperation; human expert experience and machine intelligence complement each other to improve human-computer cooperation efficiency and continuously improve the performance of intelligent information scheduling system. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 A flowchart of a method for constructing an intelligent information scheduling operation scene cognitive model is proposed. DETAILED DESCRIPTION

[0031] In order to make the technical means, creative features, purposes and effects of the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0032] Please refer to the accompanying Figure 1 A method for constructing an intelligent information scheduling operation scene cognitive model includes the following specific steps:

[0033] S1: Multi-modal perception data acquisition: Synchronously acquire multi-modal perception data (including but not limited to visual sensor, auditory sensor, environmental sensor, business system interaction log recording device) with time and space stamps related to intelligent information scheduling (covering multi-dimensional content of object state information, environmental state information and scheduling operation interaction information in the operation scene), time and space stamps to ensure the alignment and correlation of data in time and space dimensions, providing a basis for subsequent fusion processing;

[0034] S2: Feature fusion and knowledge representation: Input multi-modal perception data into a pre-constructed neural-symbol reasoning network, which is composed of a neural network module and a symbolic reasoning module, respectively used to extract deep features of each modal data and use domain knowledge graph and logical rules to abstract and fuse the extracted deep features, generate structured operation scene knowledge representation (including key elements such as scene objects and their relationships, scheduling operation constraint conditions, scene dynamic evolution law, realizing the transformation from perception data to inferable knowledge, providing a semantic-rich knowledge base for the cognitive model);

[0035] In the feature fusion and knowledge representation process, the neural symbol reasoning network adopts a hierarchical fusion strategy, first extracts and preliminarily fuses the features in the same modality to generate the same modality feature representation, then deeply fuses the different modality feature representations through cross-modality feature association and semantic mapping mechanism, and finally forms a comprehensive operation scene knowledge representation, ensuring effective integration and complementary use of different modality information;

[0036] S21: neural network module feature extraction:

[0037] S211: convolutional neural network (CNN) feature extraction (taking image data as an example):

[0038] Parameter definition: is the weight matrix of the lth layer convolution kernel, with a size of where k h is the height and width of the convolution kernel, w is the number of input channels, is the number of output channels; is the bias term of the lth layer convolution layer, with a size of w fc is the weight matrix of the fully connected layer, with a size of [n in ×n out ], n in is the number of input neurons, and n out is the number of output neurons; b fc is the bias term of the fully connected layer, with a size of [1×n out ];

[0039] Calculation process:

[0040] Convolution layer calculation: for the input image x (img) , the output feature map of the lth layer convolution layer is where x (img，l-1) is the output of the (l-1)th layer (the input image when the first layer); σ is the activation function, such as ReLU: σ(x) = max(0,x); "*" represents convolution operation;

[0041] Pooling layer calculation: if maximum pooling is used, the output feature map is: that is, divide into windows and take the maximum value of each window;

[0042] Fully connected layer calculation: after flattening the output of the pooling layer, the output feature vector f (img) of the fully connected layer is: L represents the last pooling layer; ​​

[0043] S212: Long Short Term Memory Network (LSTM) feature extraction (take sequence data as an example):

[0044] Parameter definition: w ih , w fh , w oh are input gate, forget gate, and output gate weight matrix respectively; w ch , w hh are hidden gate weight matrix; b i , b f , b o are input gate, forget gate, and output gate bias terms respectively; b c is the bias term of the cell state candidate value; h T represents the hidden state of the LSTM network at the last time step T under the condition that the sequence length is T

[0045] Calculation process:

[0046] For input sequence (T is the sequence length), the hidden state h t and the cell state c t of the LSTM are updated as follows:

[0047] Input gate:

[0048] Forget gate:

[0049] Output gate:

[0050] Cell state candidate value:

[0051] Cell state update:

[0052] Hidden state update: h t = o t ⊙tanh(c t )

[0053] Finally, the sequence feature vector f (seq) = h T ;

[0054] S22: Symbolic reasoning module knowledge representation:

[0055] S221: Domain knowledge graph construction: Construct a domain knowledge graph G=(E,R), where E is the entity set and R is the relationship set; Each entity e∈E and relationship r∈R is represented as a vector to represent its semantic embedding;

[0056] Parameter definition: is the embedding vector of entity i, dimension d e ; is the embedding vector of relation j, dimension d r ; is the m-th modality feature projection matrix, is the modality neural network output feature dimension;

[0057] S222: Symbolic inference rule definition: define symbolic inference rules, such as rule-based logical inference; suppose there is a rule: if entity e1 and e2 have a relationship r, then entity e3 can be derived, denoted as

[0058] S223: Feature fusion and knowledge representation calculation:

[0059] Parameter definition: is the feature vector extracted by the m-th modality neural network; is the feature fusion weight matrix, d sf is the fused feature dimension, d k is the knowledge graph embedding dimension; is the feature fusion bias term; is the attention weight vector of the m-th modality in feature weighted fusion;

[0060] Calculation process:

[0061] Feature projection to semantic space: project each modality feature f (m) to semantic space through projection matrix to obtain symbolic feature representation

[0062] Feature weighted fusion: weighted fusion of each modality symbolic feature, calculate the fused feature f sf : where σ is the activation function (such as ReLU);

[0063] Knowledge graph embedding fusion: further fuse the fused feature f sf with the knowledge graph embedding k g (such as the embedding combination of key entities and relationships in the operation scene) to obtain the final knowledge representation k final :

[0064] Through the above steps, the feature fusion and knowledge representation based on neural symbolic reasoning are completed, and the multi-modal perception data is converted into structured operation scene knowledge representation, providing a basis for subsequent uncertainty quantification and cognitive depth enhancement operations.

[0065] S3: Uncertainty quantification and cognitive depth enhancement: For the generated operational scenario knowledge representation, build an uncertainty quantification model to calculate the uncertainty metrics of each element and its relationship in the operational scenario knowledge representation (including multi-dimensional quantification indicators of probability uncertainty, fuzziness and inconsistency, etc.), and then guide the optimization algorithm to adjust the neural-symbolic reasoning network parameters and knowledge graph structure (considering the statistical characteristic constraints of data-driven and the logical constraints of domain knowledge guidance) to gradually reduce the uncertainty quantification indicators, so as to enhance the fine understanding and deep cognitive ability of the cognitive model to the operational scenario, make the model more accurately capture the potential laws and key details in the scenario, and realize the iterative improvement of cognitive depth;

[0066] When calculating the uncertainty metrics, the uncertainty quantification model combines the dynamic characteristics of the scene, adopts a time sequence sliding window mechanism to analyze and track the uncertainty of the scene knowledge representation at different time points, so as to more accurately capture the dynamic uncertainty change law of the scene, provide more accurate quantification basis for the entropy reduction guided optimization process, and enhance the cognitive adaptation ability of the cognitive model to the dynamic evolution of the scene;

[0067] S31: Uncertainty quantification calculation: define the uncertainty quantification indicators:

[0068] S311: Shannon entropy: used to measure the uncertainty of a random variable, the calculation formula is: Where: H(X) is the Shannon entropy of random variable X, p(x i ) is the probability of X taking the i-th value;

[0069] S312: Mutual information: measures the degree of mutual dependence between two random variables, defined as: Where: I(X;Y) is the mutual information of two random variables X and Y, p(x i ,y j ) is their joint probability distribution, p(y j ) is the probability of Y taking the j-th value, H(Y) is the Shannon entropy of random variable Y, and H(Y|X) is the conditional entropy of Y given X;

[0070] S313: Cognitive uncertainty: measured by the posterior distribution of the Bayesian model, representing the uncertainty of the model parameters, the calculation formula is: Y ep = H[P(Y|X,D)]-E Θ|D [H[Q(Y;f(X,Θ))]], where: U ep is the cognitive uncertainty metric, Y is the predicted variable, X is the input variable, D is the training data, Θ is the model parameter, and f(X,Θ) is the model prediction function; E Θ|Ddenotes the expectation of the posterior distribution P(Θ|D) of the model parameters Θ, which is calculated based on Bayes' theorem, combining the prior distribution P(Θ) and the likelihood function P(D|Θ), and reflects the probability distribution of the model parameters Θ given the data D, E Θ|D [H[Q(Y; f(X, Θ))]] = ∫ Θ P(Θ|D)·H[Q(Y; f(X, Θ))]dΘ, where P(Θ|D) is the posterior distribution of the model parameters Θ, and H[Q(Y; f(X, Θ))] is the conditional entropy of the predicted output Y under the model parameters Θ.

[0071] S32: Entropy reduction optimization process:

[0072] S321: Constructing optimization objective function: construct an optimization objective function with the goal of maximizing cognitive depth and minimizing uncertainty: where α and β are weight coefficients used to balance the influence of cognitive uncertainty and Shannon entropy in the optimization process.

[0073] S322: Parameter update rule: use gradient descent method to update the model parameters Θ:

[0074] where η is the learning rate, is the objective function the gradient of the model parameters Θ;

[0075] S33: Cognitive depth enhancement:

[0076] S331: Feature weighted fusion: the feature weighted fusion in step S223;

[0077] S332: Knowledge graph embedding fusion: the knowledge graph embedding fusion in step S223;

[0078] S34: Dynamic adaptability improvement:

[0079] S341: Time series sliding window mechanism: use a time series sliding window mechanism to analyze and track the uncertainty of the scene knowledge representation at different time steps. For time series t1, t2, …, t T , at each time step t, consider the historical data with a window size of w, and calculate the uncertainty quantification index within the current window.

[0080] S342: Uncertainty dynamic adjustment: dynamically adjust the model parameters and knowledge graph structure according to the uncertainty quantification index within the time series sliding window to adapt to the dynamic changes of the scene. For example, when a significant increase in uncertainty is detected, increase the attention weight of the model to new data to quickly adapt to changes in the scene.

[0081] Through the above steps, the uncertainty quantification and cognitive depth enhancement method based on entropy reduction guidance can effectively improve the performance of the intelligent information scheduling operation scene cognitive model, and realize the paradigm shift from "passive response" to "active cognition".

[0082] S4: Human-machine collaboration framework construction: build a human-machine collaboration framework to realize the two-way alignment of machine cognitive models and human scheduling expert cognition. On the machine side, the internal state, reasoning process and decision basis of the cognitive model are visualized and transformed into interpretability, so that human experts can intuitively understand the cognitive logic of the machine. On the human side, an interactive feedback mechanism is designed to collect feedback information such as human experts' cognitive judgment of the operation scene, experience knowledge, and comments on machine decisions, and transform these feedback information into formalized knowledge representation, which is fed back to the neural-symbolic reasoning network and uncertainty quantification model, forming a closed-loop interaction of human-machine cognition. Through continuous two-way collaboration and knowledge sharing, the efficiency of human-machine collaboration is improved, the experience advantage of human experts and the cognitive advantage of machines are fully utilized, and the performance and adaptability of the intelligent information scheduling operation scene cognitive model are optimized; thus a complete operation scene cognitive model is constructed.

[0083] In the human-machine collaboration framework, a human-machine collaborative optimization mechanism based on reinforcement learning is designed. According to the reward signals fed back by human experts and the performance indicators of machine cognitive models, the human-machine collaboration mode and knowledge fusion strategy are dynamically adjusted to encourage both sides to learn from each other and complement each other's advantages in the collaboration process, continuously improve the overall performance and generalization ability of the intelligent information scheduling operation scene cognitive model, and realize the continuous optimization and improvement of the efficiency of human-machine collaboration.

[0084] S41: Visualization and interpretation of machine cognitive model:

[0085] S411: Internal state visualization:

[0086] Feature representation visualization: the features f (m) or fused features f sf are processed by dimensionality reduction (such as using PCA or t-SNE algorithm) for visualization in 2D or 3D space;

[0087] Decision logic visualization: for symbolic reasoning-based knowledge representation k final , it is converted into a graph structure (such as a knowledge graph), which represents entities and relationships through nodes and edges, and displays the reasoning path;

[0088] S412: Interpretability transformation: feature importance explanation, calculate the contribution value of each feature to the prediction result, such as using SHAP (SHapley Additive explanations) value:

[0089]

[0090] where M is the number of features, f(S) is the model output corresponding to the feature set S, is the factorial, and is the intersection;

[0091] S42: Collection and formalization of human expert feedback:

[0092] S421: Collection of feedback information:

[0093] Cognitive judgment collection: Through the human-computer interaction interface, the cognitive judgment of human experts on the operation scene is collected, such as scene classification, key entity identification, etc.

[0094] Experience knowledge collection: Collect the experience knowledge of human experts, such as new symbolic reasoning rules or modification of existing rules;

[0095] S422: Formalization of feedback information:

[0096] Cognitive judgment formalization: Convert the cognitive judgment of human experts into a structured data format, such as label vector y human ∈{0,1} C (C is the number of categories);

[0097] Experience knowledge formalization: Convert experience knowledge into symbolic reasoning rules, such as new rules

[0098] S43: Two-way alignment and collaborative optimization of human-machine cognition:

[0099] S431: Machine cognition model update:

[0100] Feedback-based parameter update: Integrate the feedback information of human experts into the loss function of the machine learning model, such as: where: is the original loss function, is the loss term based on human feedback (such as cross-entropy loss), and λ is the weight coefficient;

[0101] Symbolic knowledge graph update: Add new symbolic reasoning rules provided by human experts to the knowledge graph G, and the updated knowledge graph G' contains new entities and relationships;

[0102] S432: Human-machine interaction reinforcement learning:

[0103] Reward signal definition: Define the reward signal r human , when the output of the machine cognition model is consistent with the feedback of human experts, give positive reward, otherwise give negative reward;

[0104] Policy update: Update the human-machine collaboration policy π using reinforcement learning algorithms (such as Q-learning or policy gradient methods): Q(s, a) = Q(s, a) + α[r human + γmax a′ Q(s', a') - Q(s, a)], where s is the state, a is the action, α is the learning rate, and γ is the discount factor;

[0105] S44: Human-machine collaboration efficiency evaluation and iterative optimization:

[0106] S441: Collaboration efficiency evaluation indicators:

[0107] Cognitive consistency evaluation: Calculate the consistency of the machine cognitive model output y machine with the human expert judgment y human , such as using accuracy (Accuracy) or F1 score: where: is the indicator function, and N is the sample size;

[0108] Decision quality evaluation: Evaluate the quality of human-machine collaboration decisions through simulation or actual operation, such as task completion time, resource utilization, etc.

[0109] S442: Iterative optimization strategy: Adjustment based on evaluation results: Adjust the machine cognitive model parameters and human-machine collaboration strategy based on the cognitive consistency and decision quality evaluation results, such as increasing the feedback weight λ or optimizing the reinforcement learning hyperparameters α, γ.

[0110] Through the above steps, a cognitive bidirectional alignment human-machine collaboration framework is constructed, realizing effective collaboration between the machine cognitive model and the human scheduling expert, and improving the human-machine collaboration efficiency.

[0111] S5: Intelligent information scheduling strategy generation and dynamic adjustment: Use the constructed cognitive model to perform real-time cognition and prediction analysis on the intelligent information scheduling operation scene, generate corresponding scheduling strategy suggestions based on the cognitive results, and dynamically adjust the scheduling strategy based on the continuous monitoring and feedback evaluation of the scheduling execution process by the cognitive model, realizing the paradigm shift of the intelligent information scheduling system from "passive response" to "active cognition", improving the scientificity, accuracy and real-time performance of scheduling decisions to adapt to the complex and changing operation scene requirements;

[0112] S51: Cognitive model prediction and strategy generation:

[0113] S511: Operation scene cognition prediction:

[0114] Input scene data: After preprocessing the real-time collected multi-modal perception data, input it into the cognitive model;

[0115] Scenario-aware prediction: process the input data using the cognitive model to obtain the scenario-aware result, the output of the cognitive model can be represented as k final , which contains semantic understanding of the operation scenario, including key entities, relationships and potential operation requirements in the scenario, etc.

[0116] S512: Scheduling strategy generation:

[0117] Strategy generation model: based on the output k final of the cognitive model, use the strategy generation model (such as the policy network in deep reinforcement learning or the rule-based policy generator) to generate the scheduling strategy;

[0118] Strategy generation formula: assuming that the strategy generation model is a function π: K→A, where K is the output space of the cognitive model, and A is the executable scheduling action space; The generation of scheduling strategy a can be represented as: a = π(k final ), where: a represents specific scheduling actions such as allocating resources, adjusting task priorities, etc.

[0119] S52: Scheduling strategy execution and feedback collection:

[0120] S521: Scheduling strategy execution: execute scheduling actions, execute the generated scheduling strategy a in the actual scheduling system, such as starting tasks, allocating resources, etc.

[0121] S522: Feedback data collection: collect feedback data, collect feedback data after executing the scheduling strategy, including scheduling results (such as task completion time, resource utilization, etc.) and new multi-modal perception data;

[0122] S53: Scheduling effect evaluation and dynamic adjustment:

[0123] S531: Scheduling effect evaluation:

[0124] Define evaluation indicators: according to the scheduling target, define indicators for evaluating the scheduling effect, such as task completion rate, resource utilization, etc.

[0125] Calculate evaluation indicators: calculate the evaluation indicators E of the scheduling effect according to the feedback data;

[0126] S532: Dynamic adjustment:

[0127] Define dynamic adjustment rules: according to the changes of evaluation indicators E, define dynamic adjustment rules, such as adjusting the parameters Θ π of the strategy generation model or the parameters Θ cognitive of the cognitive model;

[0128] Parameter update formula: use gradient descent method or other optimization algorithms to update parameters, for example:

[0129]

[0130] wherein: η π and η cognitive is the learning rate, and are the loss functions of the policy generation model and the cognitive model;

[0131] S54: Reinforcement Learning Optimization:

[0132] S541: Define Reward Signal: Design a reward function based on the scheduling effect evaluation index E, design a reward function r(E), for example: r(E) = w1·E task +w2·E resource +…,

[0133] wherein w1, w2, … are weight coefficients, and E task , E resource , … are task completion rate, resource utilization rate, etc. evaluation index;

[0134] S542: Policy Optimization:

[0135] Update Policy Generation Model: Update the parameters Θ π of the policy generation model using reinforcement learning algorithms (such as Q-learning or policy gradient method) to maximize cumulative rewards;

[0136] Policy Update Formula: In the policy gradient method, parameter update can be expressed as: wherein: α is the learning rate, J(Θ π ) is the expected cumulative reward function;

[0137] S55: Iterative Optimization of Cognitive Model:

[0138] S551: Data Update: Update the training data, add new feedback data to the training data set, for further training of the cognitive model and the policy generation model;

[0139] S552: Model Retraining: Use the updated training data to retrain the cognitive model and the policy generation model to improve the performance and adaptability of the model.

[0140] Through this intelligent information scheduling strategy generation and dynamic adjustment method based on cognitive model, real-time, dynamic and efficient intelligent information scheduling can be realized, and the overall performance and adaptability of the scheduling system can be improved.

[0141] S6: Model performance evaluation and verification: A multi-dimensional evaluation index system is adopted, including but not limited to cognitive accuracy, scheduling decision effectiveness, uncertainty reduction degree, and human-machine collaboration synergy. Combined with real operation scene data and simulation test environment, the cognitive model is comprehensively evaluated and verified for its reliability and effectiveness in actual application. According to the evaluation results, the model is iteratively optimized and improved to meet the complex application requirements in the field of intelligent information scheduling.

[0142] From the above description, it can be seen that the above-mentioned embodiments of the present application achieve the following technical effects: significantly improving scheduling accuracy: through multi-modal perception data acquisition and deep fusion, combined with neural symbolic reasoning, it can comprehensively and finely perceive various information in the operation scene, accurately grasp the scene characteristics; the scheduling strategy generated on this basis fully considers the scene details and complex factors, realizes the more optimal matching of resources and tasks, effectively improves the scheduling accuracy, and reduces resource waste and task delay.

[0143] Enhancing the cognitive ability of the model: the feature extraction and knowledge representation method of neural network and symbolic reasoning is fused, the semantic information and logical relationship in the scene data are deeply mined, and a structured knowledge graph is constructed; the uncertainty quantification and cognitive optimization mechanism guided by entropy reduction further improves the depth of the model's cognition of the scene, so that the model can more accurately understand the essence of the scene, and provide a more solid cognitive basis for intelligent scheduling decision.

[0144] Improving dynamic adaptability: the time sequence sliding window and reinforcement learning mechanism are introduced, the model monitors the dynamic changes of the scene in real time, and quickly captures uncertain factors; the strategy generation and dynamic adjustment module based on the cognitive model can autonomously optimize the scheduling strategy according to real-time feedback, ensure that the system can quickly respond appropriately when facing scene mutations, maintain efficient scheduling ability, and enhance the flexibility and robustness of intelligent information scheduling.

[0145] Improving human-machine collaboration efficiency: a cognitive bidirectional alignment human-machine collaboration framework is built to realize seamless interaction between machine cognition and human expert experience; the machine visualizes its internal state for human understanding, while collecting human feedback to optimize the model, forming an efficient closed-loop collaboration; the human expert experience and machine intelligence complement each other, improving the efficiency of human-machine collaboration and promoting the continuous improvement of the performance of the intelligent information scheduling system.

[0146] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to suggest that the scope of the present application is limited to these examples; under the idea of the present application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of the present application as described above. In order to be brief, they are not provided in detail.

[0147] The present application is intended to embrace all such alternatives, modifications, and variations that fall within the broad scope of the appended claims. Accordingly, any and all such alternatives, modifications and variations should be included within the scope of the present application.

Claims

1. A method for constructing a cognitive model for intelligent information scheduling operation scenarios, characterized in that, The specific steps include the following: S1: Multimodal sensing data acquisition: Synchronously acquire sensing data related to intelligent information scheduling from multimodal sensing sources, which are marked with spatiotemporal stamps; S2: Feature Fusion and Knowledge Representation: Multimodal perception data is input into the neural symbolic reasoning network, which consists of a neural network module and a symbolic reasoning module. The neural network module is used to extract deep features of each modality data and to perform semantic abstraction and knowledge fusion on the extracted deep features using domain knowledge graphs and logical rules to generate a structured operation scenario knowledge representation. S3: Uncertainty Quantification and Cognitive Depth Enhancement: For the generated operational scenario knowledge representation, an uncertainty quantification model is constructed to calculate the uncertainty measure of each element and its relationship in the operational scenario knowledge representation. Then, guided by entropy reduction, the parameters of the neural symbolic reasoning network and the knowledge graph structure are adjusted through optimization algorithms. S4: Human-Machine Collaboration Framework Construction: Build a human-machine collaboration framework to achieve bidirectional alignment between machine cognitive models and human scheduling expert cognition. The machine will visualize and explain the cognitive model, while the human side will collect expert feedback and transform it into knowledge representation to feed back into the neural symbolic reasoning network and uncertainty quantification model, forming a closed-loop interaction between human and machine cognition and constructing a complete cognitive model of the operation scenario. S5: Intelligent Information Scheduling Strategy Generation and Dynamic Adjustment: Utilizes the constructed cognitive model to predict operational scenarios in real time, generates scheduling strategy suggestions, monitors the scheduling execution process for feedback evaluation, and dynamically adjusts the strategy.

2. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 1, characterized in that, In S1, the sensing sources include, but are not limited to, visual sensors, auditory sensors, environmental sensors, and business system interaction log recording devices; the sensing data covers multi-dimensional content including object status information, environmental status information, and scheduling operation interaction information in the operation scenario.

3. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 1, characterized in that, In S2, the knowledge representation includes key elements such as scene objects and their relationships, scheduling operation constraints, and the dynamic evolution law of the scene, realizing the transformation from perceptual data to reasonable knowledge and providing a semantically rich knowledge foundation for cognitive models.

4. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 3, characterized in that, In S2, the neural symbolic reasoning network adopts a hierarchical fusion strategy during feature fusion and knowledge representation. First, it extracts and initially fuses features from data within the same modality to generate a feature representation of the same modality. Then, it deeply fuses feature representations of different modalities through cross-modal feature association and semantic mapping mechanisms to ultimately form a comprehensive operational scenario knowledge representation, ensuring the effective integration and complementary utilization of information from different modalities.

5. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 1, characterized in that, In S3, the uncertainty measure includes multi-dimensional quantitative indicators of probabilistic uncertainty, fuzziness, and inconsistency; the adjustment process of the optimization algorithm comprehensively considers the constraints of data-driven statistical characteristics and the logical constraints guided by domain knowledge.

6. The method for constructing a cognitive model for intelligent information scheduling operation scenarios according to claim 5, characterized in that, In S3, when calculating the uncertainty quantification model, the time-series sliding window mechanism is adopted in combination with the dynamic characteristics of the scene to perform uncertainty analysis and tracking on the scene knowledge representation at different times, providing a more accurate quantitative basis for the entropy reduction-oriented optimization process.

7. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 1, characterized in that, In S4, within the human-machine collaboration framework, on the machine side, the internal state, reasoning process, and decision-making basis of the cognitive model are visualized and made interpretable, enabling human experts to intuitively understand the machine's cognitive logic. On the human side, an interactive feedback mechanism is designed to collect feedback information from human experts regarding their cognitive judgments, experiential knowledge, and suggestions for correcting machine decisions in the operational scenario. This feedback information is then transformed into formalized knowledge representations and fed back into the neural symbolic reasoning network and the uncertainty quantification model, forming a closed-loop interaction of human-machine cognition.

8. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 7, characterized in that, In S4, the human-machine collaboration framework is designed with a human-machine collaborative optimization mechanism based on reinforcement learning. According to the reward signals fed back by human experts and the performance indicators of the machine cognitive model, the human-machine collaboration mode and knowledge fusion strategy are dynamically adjusted to encourage both humans and machines to learn from each other and complement each other's strengths in the collaboration process.

9. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 1, characterized in that, Specifically, S5 includes: using the constructed cognitive model to perform real-time cognition and predictive analysis on intelligent information scheduling operation scenarios, generating corresponding scheduling strategy suggestions based on the cognitive results, and dynamically adjusting the scheduling strategy based on continuous monitoring and feedback evaluation of the scheduling execution process using the cognitive model.

10. The method for constructing a cognitive model of an intelligent information scheduling operation scenario according to claim 1, characterized in that, It also includes model performance evaluation and verification steps, adopting a multi-dimensional evaluation index system, including but not limited to cognitive accuracy, scheduling decision effectiveness, uncertainty reduction, and human-machine collaboration. Combining real-world operational scenario data and simulated testing environments, the constructed cognitive model is comprehensively evaluated and verified, and the model is iteratively optimized and improved based on the evaluation results.

Images