41 results about "Inference machine" patented technology

This invention relates to a method and system for detecting anomalies in wastewater treatment processes based on a hybrid expert model. The method includes: constructing a multimodal input vector for the wastewater treatment plant; inputting the multimodal input vector into a hybrid expert model, and performing anomaly inference through multiple expert networks selected by sparse gating to obtain anomaly prediction results; dynamically calculating the weight coefficients of each expert network based on the feature distribution of the multimodal input vector, and using the weight coefficients to perform weighted fusion of the anomaly prediction results to generate a preliminary comprehensive detection result; performing consistency verification on the anomaly prediction results, and triggering a thought chain inference mechanism for analysis if the verification fails, and updating the preliminary comprehensive detection result based on the analysis results; performing multi-level risk assessment and decision-making on the optimized comprehensive detection result, and generating a process anomaly report including risk level and disposal recommendations. This invention can reduce false alarms and false negatives, and improve the accuracy and efficiency of anomaly identification and decision-making in wastewater treatment plants.

The application discloses a social robot behavior semantic consistency detection method for time sequence analysis and belongs to the field of social account detection.The core innovation of the method is to construct a "semantic indexed time sequence interaction graph", realize the deep fusion of semantic content and topological structure, introduce an offline reasoning mechanism, utilize a pre-training language model to construct a static semantic index matrix, adopt a double-flow architecture, fuse a multi-head graph attention network and a GRU on the behavior side, inhibit noise neighbor interference through an adaptive weighting mechanism, capture the space-time behavior evolution track of nodes, utilize a Transformer and a self-attention mechanism on the semantic side to extract deep text logic, mine potential robot gangs through the construction of a homogeneity association graph, extract cluster structure features, align "behavior-semantic" feature spaces through a cross-modal interaction module, and realize high-precision and low-cost identification of social network robot accounts in combination with a multi-task joint optimization strategy.

This invention provides a method and system for 3D reconstruction of animal surfaces using a shearing robot. First, the robot arm's scanning pose is determined through a Gaussian mutual information-based scanning pose inference mechanism to ensure multi-view coverage of the target surface by the depth camera. Then, the robot arm executes motion planning to acquire point clouds of the target from multiple perspectives, and the original point clouds are filtered and denoised. Next, coarse registration is performed using feature extraction and a robust algorithm to provide initial values ​​for fine registration. Based on the coarse registration, a Fast-GICP registration algorithm based on sliding window optimization is used, combined with a bidirectional frame-by-frame progressive strategy, to jointly optimize multiple frames of point clouds within a local time window, improving global registration accuracy. Finally, the multi-view point clouds are fused to generate a continuous and complete 3D surface model, achieving high-precision and robust 3D reconstruction of the animal surface.

This invention discloses an adaptive text-image semantic alignment method and system based on nested multi-granularity representations, relating to the field of cross-modal information retrieval technology. The invention includes: dynamically selecting different numbers of tokens from image patch tokens using a visual token selection strategy, concatenating them with classification tokens, and generating nested multi-granularity visual representations via a shared Transformer decoder layer; employing a contrastive learning strategy with granularity-difficulty forced mapping during training to couple sample difficulty with the granularity space, and utilizing a confidence consensus partitioning mechanism to suppress noise; and proposing an adaptive agile inference mechanism based on prediction entropy during inference, achieving precise matching of computational resources and query complexity by progressively calculating the entropy value of the candidate set similarity distribution and dynamically deciding whether to stop early. This invention supports seamless switching between efficiency-first and performance-first modes, significantly reducing computational overhead while ensuring retrieval accuracy, providing a flexible and scalable solution for edge deployment.

The application discloses a small sample road noise diagnosis method based on a large model and double knowledge enhancement and related equipment, which can be applied to the field of intelligent networked vehicles and artificial intelligence technology. Through the construction of the original road noise signal corresponding to the multi-channel time-frequency tensor, the application inputs the convolutional neural network model, and based on the preset processing strategy, the preliminary fault type of the road noise is preliminarily judged to obtain the preliminary fault category. Then, after parallel semantic retrieval, structured triples and unstructured document blocks are obtained. Combined with the preliminary fault category and the original road noise signal, multi-source information fusion and enhanced prompt information are constructed to obtain enhanced prompt information. Based on the explicit thinking chain reasoning mechanism in the pre-trained large language model, the road noise analysis is carried out to obtain the high-explainability inference conclusion and the maintenance suggestion of the road noise analysis. According to the road noise analysis inference conclusion and the maintenance suggestion, the target road noise diagnosis report is generated. The model can be continuously optimized and iterated based on the review information, and the accuracy of the road noise diagnosis result is improved.

This invention relates to a task-adaptive few-shot ISAR target recognition method, comprising the following steps: extracting prior structural features: extracting a set of candidate structural feature channels containing morphological information from the original image; optimal selection and representation of structural features for specific tasks; structural channel selection; a structure-aware manifold feature inference mechanism: establishing local manifold relationships within the bottleneck feature subspace, and achieving self-correction of feature distribution through structural consistency constraints and geometric regularization; and adaptive classification and stability joint optimization: combining a prototype-based metric learning strategy with structural consistency constraints. This application achieves sparse selection and optimization of structural feature channels, effectively reducing feature redundancy and enhancing discriminative ability, improving the model's ability to express structural similarity and its generalization performance in small-shot scenarios, and ensuring channel activation consistency and feature stability under different viewpoints and noise interference conditions.

This application discloses a video semantic analysis method for multimodal cognitive intelligent agents, belonging to the field of computer technology. The analysis method includes: acquiring natural language instructions and the constraints contained in the natural language instructions to obtain deconstructed intentions; acquiring multiple keyframes of the video stream based on the deconstructed intentions to generate a cognitive benchmark for semantic analysis; establishing dependencies between deconstructed intentions based on a thought chain reasoning mechanism and the cognitive benchmark to obtain a task graph; acquiring the tasks allocated within each heterogeneous computing node based on the task graph and advancing the task graph; and updating the posterior confidence of subsequent events in real time based on the observation evidence set continuously acquired during the advancement of the knowledge graph, and adjusting the logical links. This solves the problems of excessive resource consumption and the resulting difficulty in timely acquisition of effective information in existing technologies.

The application discloses a community cross-domain recommendation system based on multi-modal large model alignment, and relates to the technical field of artificial intelligence recommendation. The system includes multi-modal alignment, prompt construction and teacher-student alignment modules. First, the multi-modal large model is used to convert the heterogeneous images and text data of interest points into unified structured semantic labels, realizing cross-modal feature alignment. Second, the user's auxiliary travel history is introduced and combined with the thinking chain reasoning mechanism to accurately decouple the preference differences of users in the resident mode and the tourist mode. Finally, a two-stage strategy of supervised fine-tuning and direct preference optimization is adopted to efficiently transfer the reasoning ability of the large teacher model to the lightweight student model. The application effectively solves the semantic gap and preference deviation problem in off-site recommendation, realizes real-time recommendation within seconds while ensuring high accuracy, and is suitable for online travel services and location service scenarios.

This invention discloses a remote sensing satellite network security situational awareness method and system with a dual prevention mechanism. The method includes: acquiring multi-source heterogeneous data from a remote sensing satellite ground system network; generating a structured dataset through cleaning, denoising, and standardization; recording network assets using an asset identification algorithm based on the structured dataset; and generating a risk heatmap of the network assets using a Bayesian network model and entropy weighting method; detecting abnormal traffic and behavior using a deep learning model based on the risk heatmap and the structured dataset; and generating a priority-marked list of vulnerabilities using a general vulnerability scoring standard and a threat intelligence database; and constructing a dynamic security situation map using a graph database and community discovery algorithm based on the risk heatmap and the vulnerability list. This invention improves the real-time fusion capability of multi-source heterogeneous data, enhances the dynamic attack chain reasoning mechanism, and enables collaborative decision-making for risk warning and vulnerability management.

This invention relates to the field of signal devices and early warning systems, specifically a method and system for early warning of tunnel entrance slopes based on multi-source monitoring fusion. The method includes: acquiring data through a three-dimensional monitoring network; predicting future deformation trends using a long short-term memory network model combined with displacement and rainfall factors; evaluating the slope evolution stages using finite element numerical analysis; constructing a dynamic fact base for an expert system based on the evaluation results; using an inference engine to perform pattern matching between the fact base and a rule base containing slope cutting, surcharge counterpressure, slope protection, and drainage measures to generate suggestions; activating an alarm mechanism when a graded threshold is triggered and displaying the path and suggestions using a three-dimensional visualization platform. This invention, through multi-source monitoring fusion and expert decision-making linkage, achieves a predictive alarm system with the ability to predict disaster mechanisms and intelligently generate prevention and control measures.

The application provides a TBM jamming risk intelligent early warning method and system based on knowledge-data double driving, which comprises the following steps: acquiring multi-source parameter information collected in the TBM tunneling process of TBM tunneling cases under different stratum tunneling conditions, and constructing a multi-source parameter database; processing the data in the multi-source parameter database, and calculating data-driven jamming probability based on a robust time sequence anomaly detection model for abnormal changes; extracting expert experience knowledge based on TBM jamming literature research and experience knowledge in actual construction cases, and constructing a TBM tunneling jamming event knowledge base; determining the membership degree and the reliability of the fuzzy condition, constructing a TBM tunneling jamming event rule reasoning machine, and calculating knowledge-driven jamming probability based on a fuzzy reasoning machine model; multiplying the data-driven probability and the knowledge-driven probability by using a factor multiplication method, calculating the TBM tunneling jamming probability, and obtaining the TBM tunneling jamming risk prediction result based on the knowledge-data double driving.

The application provides a knowledge and data double-driven broken mill equipment health management industry model construction method, and belongs to the technical field of artificial intelligence.In the application, firstly, broken mill equipment multi-source knowledge and operation and maintenance data are acquired, knowledge extraction, sequence reconstruction and unified vectorization processing are performed, a mixed vector with semantic alignment is formed, differences of heterogeneous data are eliminated, and standardized input is provided for a model;then, a neural engine is constructed based on a Transformer, a symbolic reasoning engine is constructed in combination with an expert rule base and forward reasoning, an interactive checkable double-engine collaborative reasoning architecture is formed, data generalization and mechanism constraint are considered;next, real-time operation and maintenance data and query input architecture are constructed, a result is generated by the neural engine, and the symbolic engine is checked, iteratively modified, and a reliable reasoning sample is obtained;finally, a health management instruction set is constructed, a low-rank adaptive fine-tuning model is adopted, RAG retrieval and thinking chain prompting are fused, and finally integrated and highly reliable broken mill equipment health management industry model construction is completed.

This invention discloses an intelligent question-answering method and system for power metering knowledge based on multimodal retrieval. The method includes constructing a multimodal knowledge graph for power metering, multimodal encoding of user questions, execution of a hybrid retrieval strategy, and answer generation and interpretation. By constructing the multimodal knowledge graph, this method transforms heterogeneous information such as equipment parameters, technical specifications, and real-time monitoring data into a structured knowledge network, achieving semantic association and unified representation of data from different modalities. Based on a hybrid retrieval strategy and a multi-hop reasoning mechanism, this method can effectively handle complex queries involving semantic association and temporal constraints.

This invention discloses a method and system for tripping analysis and control during the entire lifecycle of a gas turbine unit, belonging to the field of intelligent control of gas turbine units. It includes a first OR module, a first greater than comparison module, a first less than comparison module, a first AND module, a second AND module, a first SR trigger module, a second SR trigger module, and a second OR module. The main functions of this invention include: trend prediction and early hazard identification based on multi-source data fusion; standardized and intelligent tripping timing inversion and root cause reasoning mechanisms; and comprehensive support for full-cycle optimization decision-making, encompassing operation, protection, maintenance, and equipment lifecycle data analysis.

The application relates to the technical field of artificial intelligence, and discloses an emergency disaster reduction intelligent decision system based on a large model and reinforcement learning. The system comprises a disaster situation sensing module, an edge coding module, a knowledge enhancement module, an inference arrangement module, a collaborative decision module, a decision checking module and an emergency execution module. Multi-source disaster situation data are collected and structured modeled, knowledge enhancement input is formed in combination with a knowledge base search result; on this basis, a candidate emergency decision trajectory is generated by adopting a reasoning mechanism of a draft model and a target model collaboration, and joint optimization of historical experience and current decision is carried out through a collaborative evolution mechanism, so that strategy continuous optimization and optimization are realized; finally, multi-constraint checking is carried out on the target decision strategy, and execution is carried out. The application improves the real-time performance, executability and self-adaptive optimization capability of emergency decision.

This invention discloses a cross-domain traffic early warning method based on policy fingerprints and digital twins, belonging to the field of network traffic monitoring. The method includes: collecting cross-domain multi-source network data from historical periods and real-time runtime; extracting long-term policy features of autonomous systems based on historical data to construct a historical policy fingerprint database and a digital twin; continuously monitoring real-time network behavior based on real-time runtime data, identifying deviations in real-time network behavior relative to historical policy fingerprints, and filtering suspicious drift events using semantic reasoning mechanisms; inputting suspicious drift events into the digital twin for subsequent propagation and evolution simulation, calculating early risk scores; and dynamically correcting the policy fingerprint database and digital twin. This invention effectively alleviates the problems of delayed early warning, high false alarm rates, and difficulty in characterizing abnormal evolution processes, improving the accuracy, stability, and robustness of early warning.

This application discloses a data processing and insight method and system based on a multimodal intelligent assistant, belonging to the field of computer technology. By integrating device physical model deduction and domain knowledge graph retrieval, a dual-track coupled reasoning framework is constructed, effectively solving the problems of decoupling between evidence and the spatiotemporal evolution of the device, and the logical defocusing of knowledge retrieval in traditional diagnosis. Through heterogeneous intelligent agent collaborative data collection and real-time conflict detection, the reliability and consistency of evidence are improved. Dynamic graph fusion and traceable reasoning mechanisms provide a transparent causal attribution chain and comprehensive credibility assessment, overcoming the limitations of black-box diagnosis. The introduction of metacognitive reflection and closed-loop optimization instructions can proactively address uncertainty, achieve continuous knowledge evolution and self-improvement of diagnostic capabilities, thereby reducing the false positive and false negative rates and supporting accurate operation and maintenance decisions.

The invention discloses an intelligent network connection vehicle traffic accident knowledge base construction system, and the system comprises a large model adaptation module which is provided with a lightweight large language model facing the field of intelligent network connection vehicle accidents; the knowledge graph construction module is configured with a structured knowledge graph which is constructed by extracting entities, attributes and relationships from a multi-source corpus based on a lightweight large language model and a preset knowledge modeling rule; the knowledge base generation module is configured with a knowledge base which is generated by carrying out collaborative scheduling on the lightweight large language model and the knowledge graph based on a graph-model complementary fusion inference mechanism; the graph-model complementary fusion inference mechanism performs fusion and confidence evaluation on semantic related knowledge from the large language model and structured associated knowledge from the knowledge graph in response to the query request, and generates an inference result based on the evaluated confidence; and the user interaction module is used for receiving the query request and visually outputting a reasoning result. Meanwhile, the invention further discloses a method and equipment.

This invention discloses a substation site table review method based on multidimensional feature fusion and template inference engine. The review process is as follows: importing and preprocessing substation site table data to output a standardized substation site table record set; extracting multidimensional feature vectors from the record set; automatically generating a standard template substation site table based on the substation's structural parameters using a template inference engine; calculating a comprehensive similarity between the multidimensional feature vectors extracted from the standardized field signal record set and the template signals of the standard template substation site table; comparing the comprehensive similarity with a set threshold; and performing the substation site table review based on the comparison result. By combining a template inference engine with a multidimensional feature fusion algorithm, the entire process from substation site table import, signal feature recognition, automatic template matching, and intelligent review is automated.

The application relates to the technical field of autonomous value keeping based on a rule tree, and discloses an autonomous value keeping method and system based on a rule tree, which comprises the following steps: the autonomous value keeping method and system based on the rule tree provided by the application introduce a rule tree and a behavior logic engine, realize unmanned autonomous operation of a training station, break the dependence of traditional training on human operators, effectively reduce the idle time of the training station, and fully utilize existing resources; each operation task can be automatically executed according to a preset rule tree logic, the continuity and consistency of a training process are ensured, training interruption caused by the intermittence of manual operation is avoided, and the coherence of a training task is ensured; a dynamic rule reasoning mechanism is adopted, an operation decision path is adjusted through real-time state feedback, the judgment logic and the behavior mode of a real operator can be simulated, the system has a certain adaptability, and the intelligent level of simulation training is improved.

The present application relates to big data processing and edge computing technical field, specifically to information big data management system based on edge computing, including mutation detection module, skip list management module and decision execution module, the mutation detection module carries out "noise-silence" dual-state detection to big data stream in edge node, only intercepts mutation micro-slice at state conversion and calculates 128 bit SimHash mutation fingerprint; the skip list management module weaves the "mutation fingerprint-physical block" bidirectional skip list with reversible fuse mark on local PCM, realizes the index of mutation fingerprint to physical block address; the decision execution module takes the skip list root handle as the entrance, drives the "write once" type edge inference machine to generate the immutable decision frame, writes back the data source terminal through the low delay channel and updates the fuse state, so as to complete the closed loop processing of cold mutation data.

This invention discloses a method and system for querying DWG drawing text information based on a multimodal large model. The method relates to the field of DWG drawing text information query technology and includes the following steps: data parsing and multimodal representation construction, multimodal large model semantic understanding and reasoning, structured output and post-processing, and backup output assurance. This invention parses DWG files, extracts text and its coordinates, generates a structured description, inputs this description and user query commands into a multimodal large model for semantic matching and reasoning, performs structured transformation, logical verification, and knowledge base association on the output results, and finally generates the query results. When text parsing fails, a backup reasoning mechanism of the large model is activated to ensure system availability, improving the adaptability of DWG drawing text information query and solving the problem of low adaptability in existing technologies.

This invention discloses an incomplete cross-modal hash retrieval method based on predictive completion and auxiliary code guidance, belonging to the field of cross-modal hash retrieval technology. This invention employs a pre-trained CLIP model as a cross-modal encoder to extract semantic features from image and text modalities, effectively enhancing the semantic consistency of cross-modal data. A bidirectional predictive network is designed, combined with a variational inference mechanism, to efficiently complete missing modal features using available modal features, successfully addressing the problem of modal loss caused by data acquisition failures or transmission interruptions. A residual contrastive network is introduced to improve the discriminative ability of the completed multimodal features and achieve feature alignment, effectively mitigating the distribution shift caused by missing data. In the hash encoding stage, a high-bit auxiliary code is used as a semantic teacher, guiding the low-bit hash code to learn richer semantic representations through knowledge distillation, significantly reducing information loss during compression. This invention significantly improves the efficiency and accuracy of incomplete cross-modal retrieval.

This application discloses an autonomous representation processing method, system, device, and storage medium, relating to the field of artificial intelligence technology. The method includes: receiving user-input representation requirements; analyzing and decomposing the representation requirements using a large language model and a pre-set thought chain reasoning mechanism to obtain a task flow; performing imaging and positioning operations according to the task flow to acquire imaging data; extracting target object instances and calculating attributes from the imaging data to generate instance-level structured objects; and forming representation conclusions and / or decision information or physical execution instructions based on the instance-level structured objects. This method utilizes a large model to achieve autonomous task planning, and parallel visual analysis supports automatic positioning, filtering, measurement, and statistics, significantly reducing manual costs. Dynamic process decomposition improves individual recognition accuracy in complex scenarios. Executable instructions are generated based on pixel-physical mapping, establishing a closed loop from image perception to physical operation, effectively solving the problem of cross-device collaboration.

The application relates to a three-dimensional model intelligent assembly design and verification method, in particular to the artificial intelligence field, through unified feature extraction and cross-modal semantic alignment of multi-source heterogeneous three-dimensional data, deep understanding of the geometric shape and functional intention of parts is realized, an assembly knowledge graph and a relationship perception reasoning mechanism constructed based on historical experience can intelligently infer potential assembly relationships and constraint parameters of new parts, more importantly, a dynamic closed loop formed by virtual verification feedback drives continuous expansion of the assembly knowledge graph and continuous optimization of semantic representation, so that the assembly design automation degree and accuracy are improved, the adaptation and differentiation ability to new parts and easily confused parts are significantly enhanced, and a complete intelligent cycle from perception, reasoning to self-evolution is realized.

The application discloses a medical insurance audit intelligent agent and a medical insurance audit management system, and belongs to the technical field of medical insurance audit, and specifically can realize the following steps: obtaining multi-source heterogeneous data and performing data cleaning to obtain document data; based on a medical insurance audit knowledge base, meta data related to rules is screened and summarized from the text data; a large language model performs reasoning based on audit prompt words integrated with the medical insurance audit knowledge base and the meta data and a CoT reasoning mechanism, and outputs an audit result. The medical insurance audit intelligent agent is used for implementing document quality control on medical records in the diagnosis and treatment process, and is also used for the audit and complaint process after medical insurance settlement, realizing real-time reminding intervention and post-complaint feedback on medical insurance settlement, and improving the accuracy and timeliness of medical insurance audit.

This invention belongs to the field of target detection and classification technology, specifically relating to a few-shot ISAR target recognition method based on TA-SCSB. The method includes the following steps: Step 1: Structural prior feature extraction stage; Step 2: Task-adaptive structural statistical modeling stage; Step 3: Structural channel selection bottleneck stage; Step 4: Determination of structure-aware manifold feature inference mechanism stage; Step 5: Adaptive classification and stability joint optimization stage. This method introduces a task-level structural statistical modeling and channel selection coupling mechanism into few-shot ISAR recognition for the first time. It dynamically generates channel weight distributions based on the structural statistical features (variance, information entropy, gradient strength) of supporting samples, achieving task-adaptive selection and weighting of feature channels.