12 results about "Web intelligence" patented technology

The invention discloses an encrypted traffic adaptive update classification method and system oriented to an open network environment. The method comprises the following steps: firstly, extracting endogenous semantic features and exogenous environment features based on a causal decoupling mechanism, and stripping an environment confusion factor through anti-fact disturbance and invariance constraint to obtain invariant semantic representation; constructing a macroscopic drift state vector representing a network situation, and inputting a trained meta-learning super-network intelligent decision adaptive control hyper-parameter; performing cross-modal element calibration by utilizing large language model thinking chain reasoning, and calculating the subspace direction consistency of an instantaneous gradient vector of a candidate sample and a category optimization trajectory prototype so as to screen credible samples; and in combination with the capacity-limited playback queue, gradient orthogonal projection constraints are introduced to update low-rank adaptation layer parameters. According to the method, the concept drift problem is solved through causal decoupling and meta-learning decision, forgetting prevention is achieved through orthogonal projection updating, and the online adaptability and robustness of the model in the open environment can be improved without additional manual annotation.

The invention discloses a cross-platform dark web information dynamic tracking method based on federated learning, and relates to the technical field of information tracking, and the method comprises the steps: a federated server carries out the weighted average aggregation of an encrypted local model parameter gradient, generates a global model updating gradient, distributes the global model updating gradient to each federated learning client, and transmits the global model updating gradient to each federated learning client; each federated learning client updates the local model after decryption to obtain a new generation of global model, and when any federated learning client finds a suspicious target, each federated learning client performs joint reasoning calculation by using the new generation of global model and the local multi-modal feature vector to generate cross-platform associated information, and the cross-platform associated information is obtained. And writing the cross-platform association information as an evidence storage record into the block chain. According to the integrated intelligence tracking method, the accuracy, the model convergence speed and the generalization ability of dark web entity association are improved, and the core pain points of multi-source information splitting, high privacy leakage risk, difficulty in intelligence solidification and the like in the conventional method are solved.

The invention provides an agent-oriented website map construction method. The method comprises the following steps: firstly, defining a design criterion of network agent website map service, then providing a website map building module based on the design criterion, and outputting the website map service which can be adopted by an LLM-driven agent; the website map building module is composed of three stages: a first stage: website structure analysis: finding a core page by traversing a hyperlink structure of a website, carrying out breadth-first search by using a crawler, controlling an exploration range by adopting a configurable scoring function, and eliminating noise through a dynamic content and navigation noise filtering algorithm; in the second stage, user exploration is simulated, Playwright is used as an automatic agent, an interactive element exploration and dynamic content triggering algorithm is designed to trigger loading of dynamic content and reveal hidden elements, and dynamic content capture or clean page snapshot is carried out; and a third stage: browsing the content annotation. The capability is enhanced by providing external knowledge, so that the computing resource and time cost of technology implementation is greatly reduced.

The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and / or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a “meta-trainer,” that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available. In either case, dense, potential-based rewards may be used to augment the training.

An application generates a first user interface to enable a plurality of prompts to be combined into one or more groups of prompts, where each prompt of the plurality of prompts represents a filter for data targeted by subsequent queries. Next, the application generates a second user interface displaying the one or more groups of prompts in response to detecting that the plurality of prompts have been combined into the one or more groups of prompts in the first user interface. Then, a web intelligence engine executes a query in response to one or more values being specified, in the second user interface, for the one or more groups of prompts, where the one or more values filter the data returned by the query. Next, the web intelligence engine returns a result of the query to a first computing device based on executing the query.

The invention discloses a fuel gas temperature control steel ladle baking control method and system based on multi-modal big data deep reinforcement learning. The method comprises the following steps that S1, multi-source heterogeneous data fusion modeling is carried out, and feature level fusion is carried out; s2, training a depth prediction model and outputting a result: establishing a baking process prediction model; s3, constructing a migration reinforcement learning agent: establishing a dual-channel deep reinforcement learning architecture, including establishing an offline pre-training channel and an online fine tuning channel; and S4, virtual-real collaborative optimization control: collecting multi-modal data in real time, filtering noise samples, updating model parameters, finely adjusting a strategy network, and outputting an air-fuel ratio adjustment decision by an intelligent agent. According to the method, the deep reinforcement learning model driven by enterprise-level real production data is constructed, the air-fuel ratio is dynamically regulated and controlled, three-dimensional cooperative control of energy efficiency optimization, quality control and environmental protection indexes is achieved, and the technical bottlenecks that in a traditional control method, simulation data adaptability is poor, and single-target adjustment is limited are solved.

This invention discloses a distributed flexible job shop intelligent scheduling method oriented towards process-bound constraints, belonging to the field of computer vision technology. The method constructs a distributed scheduling environment simulator and constraint manager to output a normalized three-dimensional state feature vector; it filters the legal action set through constraint verification; it then constructs an improved duel depth Q-network agent to separate state value and action advantage flow to improve estimation accuracy; it introduces a priority experience replay mechanism to accelerate convergence, employs an ε-greedy strategy combined with action masking to ensure action validity, and triggers a heuristic backoff strategy when no valid action is found; it iteratively trains the agent guided by a composite reward function to achieve multi-objective optimization of completion time, load balancing, and compliance with process-bound constraints; finally, it extracts the optimal scheduling scheme from non-dominated solutions. This invention overcomes the shortcomings of traditional algorithms in constraint handling and susceptibility to local optima, achieving a synergistic improvement in constraint satisfaction and multi-objective optimization.

This invention discloses an intelligent operation and maintenance (O&M) and emergency response system and method for communication networks. The system comprises an intranet large-scale model platform module, a local API service module, a GNN topology inference module, a standardized database module, an interface adaptation and streaming transmission module, an AI agent autonomous execution module, a data integrity and self-consistency verification module, and an intelligent O&M and emergency expansion module. Based on intranet large-scale model and graph neural network technology, this system constructs an intelligent O&M system for communication networks, achieving rapid and accurate handling of all types of faults, real-time controllable O&M data, secure implementation of intranet AI, and automated closed-loop O&M. This improves network O&M efficiency and stability, ensures core data security, and is suitable for industrial-grade high-security and high-intensity O&M scenarios.

The application belongs to the technical field of communication network operation and management, and discloses an AI-based communication network intelligent inspection and maintenance system, which interacts with the Web interface of a multi-manufacturer network management, carries out document object model analysis on a page, establishes an interface operation model irrelevant to manufacturer protocols, configures inspection items according to network element types and service scenarios, and controls a browser to complete cross-page operation to collect running data. The system fuses inspection data with historical data, work orders and equipment history, constructs a running state view organized according to network elements and time, and identifies abnormal types and levels by combining rules and classification models. Furthermore, the system generates a maintenance strategy containing inspection and change steps by relying on a maintenance operation template library, executes a maintenance task by combining test session simulation and rollback control, and improves automation and running reliability.

The application discloses a cross-platform dark web intelligence dynamic tracking method based on federal learning, relates to the technical field of intelligence tracking, and comprises the following steps: a federal server performs weighted average aggregation on encrypted local model parameter gradients, generates a global model update gradient, distributes the global model update gradient to each federal learning client, each federal learning client updates a local model after decryption, and obtains a new generation of global model; when any federal learning client finds a suspicious target, each federal learning client uses the new generation of global model and a local multi-modal feature vector to perform joint reasoning calculation, generates cross-platform related intelligence, and writes the cross-platform related intelligence into a blockchain as a notarization record. The application improves the accuracy of dark web entity correlation, the model convergence speed and the generalization ability through an integrated intelligence tracking method, and also solves the core pain points in the previous methods, such as multi-source information fragmentation, high privacy leakage risk and difficult intelligence solidification.

This application discloses a method, apparatus, device, and storage medium for generating access processes based on semantic graphs, relating to the field of process management technology. The method includes: acquiring access requirement information, a preset semantic graph model, and a preset process template. The access requirement information includes user access type, access location, authentication method, and security policy; determining a multi-dimensional semantic graph based on the access requirement information and the preset semantic graph model, wherein the multi-dimensional semantic graph consists of semantic nodes and semantic edges, with semantic nodes having node attributes including weight and confidence scores, and semantic edges having edge attributes including cost functions; determining a process path based on the multi-dimensional semantic graph; and mapping the process path into the preset process template to obtain the target access process. This achieves the generation of controllable and traceable target access processes, thereby improving the automation capabilities and operational efficiency of intelligent network access systems.

The invention belongs to the technical field of communication network operation, maintenance and management, and discloses an AI-based communication network intelligent inspection and maintenance system, which interacts with a Web interface of a multi-manufacturer network manager, performs document object model analysis on a page, establishes an interface operation model irrelevant to a manufacturer protocol, configures an inspection item according to a network element type and a service scene, and provides an intelligent inspection and maintenance system for a communication network. And the browser is controlled to complete cross-page operation and collect operation data. The system fuses inspection data with historical data, work orders and equipment resumes, constructs a running state view organized according to network elements and time, and identifies the types and grades of abnormities in a manner of combining rules with a classification model. And a maintenance strategy including checking and changing steps is generated by relying on the maintenance operation template library, and a maintenance task is executed by combining test session simulation and rollback control, so that the automation and operation reliability is improved.