5 results about "Markov logic network" patented technology

The invention is suitable for the technical field of artificial intelligence and multi-modal reasoning, and provides a neural symbol reasoning method and system for multi-modal information processing, and the method comprises the following steps: obtaining multi-modal input data; respectively extracting semantic attribute representations of the image data and the text data on a predefined attribute set to obtain a multi-modal semantic attribute representation; mapping the multi-modal semantic attribute representation into a predicate in a first-order logic form, and constructing a symbolized multi-modal attribute fact set; inputting the multi-modal attribute fact set and a logic rule in a pre-constructed first-order logic knowledge base into a Markov logic network for reasoning to obtain a reasoning result; and integrating the reasoning results, and generating an interpretable decision result and a reasoning logic link. According to the method, multi-modal shared attribute symbols are taken as bridges, information fusion, conflict resolution and high-level consistency constraint among multiple modals are realized, and a reasoning result with interpretability, high robustness and strong generalization ability can be obtained.

The invention relates to an intelligence knowledge graph completion method based on knowledge and data dual drive, and belongs to the field of knowledge graph reasoning. The method comprises the steps of performing intelligence entity extraction and intelligence relationship extraction based on acquired intelligence data to obtain an entity relationship triple representing intelligence facts, and constructing an intelligence knowledge graph; calculating to obtain an approximate posterior distribution probability of each intelligence fact by constructing a fact establishment possibility evaluation function; constructing a Markov logic network, and establishing a joint probability distribution model of known information facts and missing information facts; and improving an optimization function of the variational reasoning calculation process based on the approximate posterior distribution probability, and solving the joint probability distribution model based on the step E and the step M of the variational reasoning calculation process to obtain missing intelligence. According to the method, logic rules and neural network embedded representation are combined, efficient reasoning is achieved, and complex and diversified intelligence reasoning tasks can be conducted on a large-scale intelligence knowledge graph.

This invention relates to the fields of artificial intelligence and multimodal reasoning technology, and provides a neural symbolic reasoning method and system for multimodal information processing. The method includes the following steps: acquiring multimodal input data; extracting semantic attribute representations of image data and text data on a predefined attribute set to obtain multimodal semantic attribute representations; mapping the multimodal semantic attribute representations to predicates in first-order logic form to construct a symbolic multimodal attribute fact set; inputting the multimodal attribute fact set and logical rules from a pre-constructed first-order logic knowledge base into a Markov logic network for reasoning to obtain a reasoning result; integrating the reasoning result to generate an interpretable decision result and a reasoning logic chain. This invention uses shared attribute symbols across multiple modalities as a bridge to achieve information fusion, conflict resolution, and high-level consistency constraints among modalities, resulting in reasoning results with interpretability, high robustness, and strong generalization ability.

The invention discloses a photovoltaic facility operation and maintenance decision-making method based on a knowledge graph, and the method comprises the steps: S1, constructing the knowledge graph, and building equipment, fault modes, feature parameter nodes and causal association; s2, performing state diagnosis based on an improved Markov logic network, and obtaining an equipment state probability, a fault propagation chain and an influence degree by using a structure regularization factorization model and a confidence degree adjustment mechanism; s3, constructing an operation and maintenance index system, extracting risks, fault influences and performance degradation, and forming an evaluation vector in combination with the cost and the power generation recovery rate; s4, optimizing an operation and maintenance strategy based on an improved VIKOR method, introducing an index coupling degree matrix and a compromise correction factor, and calculating group utility, individual regret and a compromise evaluation value; and S5, outputting an operation and maintenance decision, generating priority and alternative schemes, and providing a basis for fault chain interpretation and weight correction. Intelligent diagnosis and operation and maintenance auxiliary decision making of the photovoltaic facility are realized, the operation and maintenance efficiency is improved, the fault risk is reduced, and the power generation income is optimized.

The present disclosure relates to a method, device and medium for performing retrosynthesis analysis on a compound molecule. A method for performing retrosynthesis analysis on a compound molecule is provided, comprising: determining probability graph weights of single-step reactions of a Markov logic network based on actual reaction learning data, thereby obtaining a trained Markov logic network; receiving an encoded molecular structure of a target compound; inputting the encoded molecular structure into the trained Markov logic network, and predicting a plurality of candidate retrosynthesis paths of the target compound through the Markov logic network; inputting the plurality of candidate retrosynthesis paths into a reinforcement learning model, training the reinforcement learning model based on existing actual reaction processes, so that a path consistent with an actual synthesis path obtains a reward, and updating a transition matrix in the reinforcement learning model; and determining one or more selected retrosynthesis paths from the plurality of candidate retrosynthesis paths by using the finally obtained transition matrix through the reinforcement learning model.