655 results about "Scale model" patented technology

The invention relates to the technical field of low-altitude safety emergency, and discloses a low-altitude safety emergency system based on large model application. The system comprises a low-altitude situation awareness module which constructs a dynamic situation prediction model based on historical flight data, collects a low-altitude target track coordinate sequence, radar echoes and communication signal characteristics in real time, and outputs a theoretical security situation value. The multi-source data fusion analysis module performs three-dimensional difference analysis on theoretical and actually measured situation values, which relates to time domain trend deviation, space domain coverage offset and event sequence matching degree, and generates a target-level difference feature matrix; and the risk propagation path deduction module inputs the matrix into a geographic topology analysis network, generates a probability thermal distribution diagram of an abnormal risk propagation path in combination with airspace node communication parameters and a target position, and locates a risk influence area. And the intelligent disposal strategy generation module configures parameters according to the graph, starts high-frequency monitoring for a high-probability area, and performs communication disturbance test on an adjacent target.

The invention discloses a spatial registration method for tunnel multi-source heterogeneous data, and the method comprises the steps: constructing a lightweight digital twinborn model based on a design parameter library, a construction log library and a real-time monitoring library; mapping the multi-source sensor data in the real-time monitoring library to a digital twin model space node through a space-time synchronization algorithm; probability distribution model-based abnormal value elimination is performed on the geological exploration data, the while-drilling parameters and the geological sketch, and a discrete smooth interpolation algorithm is adopted to generate a surrounding rock mechanical parameter three-dimensional distribution field; the surrounding rock mechanical parameter three-dimensional distribution field serves as input, and front surrounding rock classification probability distribution is dynamically predicted through a machine learning model optimized through transfer learning; based on the three-dimensional point cloud data of the tunnel face in the digital twin model, extracting geometric parameters of the structural plane and constructing a fracture network model constrained by a normal vector; according to the method, the surrounding rock risk location is accurately positioned based on double-threshold triggering, and multi-scale model seamless fusion is realized through gradient constraint topology fusion.

The invention provides a large model autonomous evaluation method, device and equipment based on a multi-agent technology, and a storage medium. The method comprises the following steps: generating an evaluation task set covering multiple scenes and grading multiple difficulties by setting dimension weights of basic indexes and extension indexes; a multi-agent evaluation system of a questioner, an evaluator and a reference agent is constructed based on the task set, a parallel interaction mechanism is adopted to simultaneously initiate tasks to a to-be-evaluated model and the reference agent, the evaluator agent compares the to-be-evaluated model and the reference agent according to a multi-dimensional scoring standard and outputs the two agents, and task instances, output results and scores are constructed into a structured data set; the subsequent task set is dynamically optimized through analysis of the data set, comprehensive and iterative evaluation of the capacity of the large model is achieved, and the problem that an existing evaluation mode cannot comprehensively reflect the real capacity level of the large model is solved.

The invention relates to the technical field of nuclear power internal disaster analysis, and discloses a nuclear power plant internal flooding experiment verification method, device, equipment and medium, and the method comprises the steps: carrying out the modeling analysis of a target plant prototype, and determining a plurality of model parameter values of an experiment scaling model; constructing an experiment scaling model according to the plurality of model parameter values, and determining experiment parameters of an internal water logging experiment; performing an internal water logging experiment on the experimental scale model by using the experimental parameters to obtain an internal water logging experiment result of the experimental scale model; and obtaining an internal water logging simulation result of the numerical model, comparing the internal water logging simulation result with an internal water logging experiment result, and verifying the accuracy of the internal water logging simulation result according to a comparison result. According to the method, the experiment scale model of the nuclear power plant prototype is constructed, the flooding spreading conditions under different working conditions are simulated and compared with the flooding simulation result, the accuracy of flooding simulation calculation is verified, and more accurate guidance is provided for follow-up nuclear power project internal flooding analysis.

The invention discloses a knowledge-based visual question-answering method and device based on big and small model collaboration and a medium, belongs to the technical field of visual question-answering, and solves the problem of how to improve the accuracy of complex questions faced by a visual question-answering technology. In the image description extraction step, an image description and key object label set which is highly associated with a natural language problem is generated, and in the entity enhancement processing step, key entities are extracted and a clarity problem and an entity example which are used for resolving semantic ambiguity are generated. In the candidate answer generation step, a plurality of semantically complementary answers are generated under different reasoning dimensions on the basis of a multi-round generation strategy, optimized candidate answers are screened on the basis of a model built-in scoring mechanism, in the context example retrieval step, examples most related to current input are retrieved, and finally, the candidate answers are obtained. And the large language model receives a unified input prompt formed by splicing output results in the steps, and inference is performed in an autoregression mode to generate answers, so that the accuracy of the visual question-answering technology facing complex questions is effectively improved.

The invention belongs to the field of automobile evaluation, and provides an automobile live risk evaluation system and method based on a large model, and the method comprises the steps: receiving voice input information of an evaluator in a vehicle on-site inspection process; extracting component state information of the target vehicle based on the structured check item text; acquiring historical maintenance record data of the target vehicle, and performing association fusion on the historical maintenance record data and the component state information; based on a large language model, performing joint reasoning on a currently input inspection label and the historical maintenance record, constructing a causal chain related to an accident risk, and generating at least one subsequent to-be-inspected item; generating voice guide content and feeding back the voice guide content to the evaluator so as to prompt the evaluator to continue to check the target item of the target vehicle; gradually perfecting the risk check path of the target vehicle until the risk chain closed loop is completed; and calling an estimation formula to calculate a residual estimation value of the target vehicle, and outputting a structured risk assessment report.

The invention discloses a method and a system for predicting the lumpiness of open-air throwing blasting, and relates to the technical field of rock blasting engineering. The invention discloses a new calculation relation about a lumpiness similarity proportion, relates to the technical field of open-air loose blasting mining, and can predict blasting lumpiness of projects such as mines, tunnels and the like by scaling a lumpiness distribution proportion. The method comprises the following steps: firstly, acquiring a rock sample from a blasting site, measuring parameters such as rock density and fracture toughness, and acquiring blasting parameter information of a to-be-blasted area; and then, a numerical simulation model under an open-air throwing blasting similar material is constructed by using a numerical simulation method, the change rule of blasting fragmentation lumpiness along with explosive unit consumption in a scale model test is studied, and the blasting lumpiness in a field blasting area is predicted according to the blasting lumpiness similarity ratio. According to the method, the change rule along with the unit consumption of the explosive is analyzed, the utilization rate of crushing energy of the explosive in surface blasting engineering is improved, reference is provided for blasting lumpiness control, and prediction of field blasting lumpiness before blasting is achieved.

The invention relates to the field of carbon fiber composite material mechanics, in particular to a fine evaluation method for the strength of a small-size large or full-opening CFRP (carbon fiber reinforced plastic) shell, which comprises the following steps: establishing a microscopic damage model containing material anisotropy, defining damage variables and evolution equations, uniformly quantifying multi-scale damage behaviors such as fiber fracture, matrix cracking and layered expansion, and evaluating the strength of the small-size large or full-opening CFRP shell. The problem that a traditional single-scale model cannot represent the multiphase coupling effect is solved; through cross-scale transmission of damage variables, whole-process prediction from microscopic damage initiation to macroscopic failure is realized, and the precision of strength evaluation of the complex opening structure is significantly improved. The method realizes refinement, reliability and engineering practicability of strength evaluation of the small-size large-opening or full-opening CFRP shell, is particularly suitable for the fields of aerospace and the like with extremely high requirements on structural weight and safety, and has remarkable technical leading and application value.

The invention discloses a wind driven generator structure high-cycle fatigue cross-platform simulation method based on multi-scale coupling modeling, and aims to accurately evaluate a fatigue damage state of a wind driven generator structure under the action of a high-cycle low-amplitude load. Firstly, a multi-scale model based on coupling of a fan overall structure finite element model unit and a local cracking area near-field dynamic model is established, and fan power time-history analysis is carried out based on a finite element analysis platform; then response data are transmitted to a processing platform, a rain flow counting method and a linear damage accumulation criterion are combined, the fatigue damage state of the connection key is determined, a fatigue life attenuation value is calculated, the fatigue cycle number needed by local crack propagation is determined, and the residual fatigue life of the unbroken key is updated; and finally, correcting the fan model, removing a newly added fracture key at the tip of the crack, and repeating dynamic time-history analysis and damage updating until the structure fails. According to the method, whole-process simulation and fatigue behavior analysis of fan fatigue crack propagation are realized, and important reference is provided for fan design optimization and maintenance.

The invention relates to the technical field of artificial intelligence, and discloses a multi-level large model scheduling and knowledge base adaptive optimization method and system, and the method comprises the steps: obtaining a to-be-processed task, carrying out the preprocessing of the to-be-processed task, and calculating the complexity of the to-be-processed task; processing the to-be-processed task by using the small model to obtain a test processing result, and calculating the maturity of the test processing result; determining a scheduling strategy of the large model according to the complexity and the maturity, and processing the to-be-processed task by using the scheduling strategy to obtain a processing result; and optimizing a small model or a knowledge base related to the scheduling strategy based on a processing result. By calculating the complexity of a to-be-processed task and testing the maturity of a processing result, models of different scales are selected accordingly, calling of a super-large model is reduced, the calculation cost is reduced, the system response efficiency is improved, a small model or a knowledge base is optimized through a historical processing result, the processing capacity of the small model and the knowledge base is improved, knowledge is continuously accumulated, and the processing efficiency is improved. And repeated calculation is reduced.

The invention relates to the technical field of wind tunnel tests, in particular to a large-scale model wind tunnel test device and a debugging method thereof. The scale model wind tunnel test device comprises a supporting device which comprises an upper supporting truss, a lower supporting truss and a plurality of supporting stand columns, the upper supporting truss and the lower supporting truss are composed of main beams and anchoring structures, the supporting stand columns are connected with the upper supporting truss and the lower supporting truss, and sliding rails are symmetrically arranged on the upper supporting truss and the lower supporting truss. The large-scale model is connected between the two sliding rails through a line spacing adjusting device and an inclination angle adjusting device, large-scale model assemblies are fixed to the upper end and the lower end of the large-scale model through detachable force transmission plates, and the force transmission plates are connected with a force measuring balance; the large-scale model is provided with a pressure measuring hole, and the scale ratio of the large-scale model is greater than 1: 20; a wind pressure sensor is arranged in the data acquisition device, the data acquisition device is connected with the pressure measuring hole of the large-scale model through a pressure measuring pipe, and wind pressure is transmitted to the wind pressure sensor of the data acquisition device through the pressure measuring pipe. The large-scale model wind tunnel test device and the debugging method thereof are convenient and fast to install and flexible to adjust.

The invention provides a matching recommendation method and system based on large-model digital talent inventory, relates to the technical field of artificial intelligence and human resource management, integrates large-model analysis, employee portrait modeling, intelligent matching recommendation and visual display of comparative analysis, has the characteristics of automation, high precision and interpretability, and is suitable for popularization and application. The problems of difficult man-post matching, inaccurate talent evaluation, opaque selection process and the like are solved, and digital talent selection and intelligent decision making are realized.

The invention relates to the technical field of data testing, in particular to a testing method and system based on large model multi-modal generation and dynamic verification, and the testing method based on large model multi-modal generation and dynamic verification comprises the following steps: initializing a testing system, deploying a pre-trained large model, and generating testing data through the large model; dynamically detecting the generated test data through a preset abnormal mode knowledge base, and feeding back an abnormal result; and according to an abnormal result, reasoning and positioning an abnormal root cause through a preset causal reasoning model, and carrying out interpretable causal analysis output. According to the method, an iteration mechanism of test data generation, detection and feedback can be established, the generation work of the test data is reversely optimized through the abnormal result, all the generated test data can be ensured to pass basic verification, the compliance of the test data is ensured, dynamic detection is performed through the preset abnormal mode knowledge base, and the test efficiency is improved. Compared with a traditional static model, the method is higher in detection accuracy and processing efficiency.

The invention belongs to the technical field of intelligent manufacturing of aeronautical manufacturing enterprises, and relates to an assembly process procedure auxiliary compilation method based on a large model, which comprises the following steps: data preparation and processing: collecting and preprocessing an assembly process procedure text file; carrying out algorithm model selection and fine tuning, using an assembly process procedure text as a training data set and a test data set, and selecting different types of large models to carry out fine tuning; service publishing: deploying a plurality of large models with the best evaluation indexes as services, publishing the services, and externally providing available interface services; constructing an RAG knowledge base, vectorizing the assembly process procedure text, and constructing a vector knowledge base; intelligent question and answer: performing knowledge question and answer based on the RAG knowledge base; and assembly process procedure file generation: constructing the assembly process procedure file through text generation. According to the method, the functions of semantic retrieval intelligent question answering of the assembly technological procedure knowledge base, automatic assembly technological procedure file generation, information recording and the like are achieved.

The invention discloses a fluid distribution dynamic regulation and control method and device based on a shale reservoir microstructure, and relates to the technical field of oil and gas exploitation. The method comprises the steps that microstructure parameters of a target shale reservoir are obtained through a core experiment, and a multi-scale model is constructed based on the microstructure parameters; determining control equations of different physical fields of the multi-scale model; coupling the control equations of different physical fields, and resolving through a finite element method; performing a numerical simulation experiment based on the multi-scale model and the resolving result to determine regulation and control parameters; the regulation and control parameters are input into the multi-scale optimization model to optimize the regulation and control parameters, and a regulation and control strategy is obtained. The problems that an existing displacement experiment is single, the complex interaction process under the actual mining environment is not comprehensively considered, and the guiding effect on mining work is limited are solved. Business personnel can be guided to formulate a development scheme more scientifically and accurately, the mining efficiency and the success rate are improved, the development cost can be reduced, and meanwhile economic benefits are improved.

The invention provides a work order generation method and device based on large model intelligent classification, equipment and a medium, and the method comprises the steps: constructing a plurality of first labels and second labels according to first-level classification, second-level classification and third-level keywords, the third-level keywords being classification subjects of classification description information; when the work order description information is obtained, candidate tags are determined from the second tags based on a classification model, and the corresponding classification description information is configured as model cue words to determine identification tags; and determining the first tag corresponding to the secondary classification included in the identification tag as a target tag, and generating a target work order based on the target tag and the work order description information. According to the technical scheme, each scene classification only corresponds to one second label, the detailed classification description information serves as additional information of the three-level keywords, the number of the classification labels is reduced, illusion classification is avoided by taking the classification description information as the model cue word, and the recognition efficiency and accuracy of the large model are improved.

The invention provides a large model evaluation method, a large model optimization method and device, electronic equipment, a storage medium, an intelligent agent and a program product, and relates to the technical field of artificial intelligence, in particular to the technical field of large models, the technical field of evaluation analysis, the technical field of optimization and the like. According to the specific implementation scheme, a dialogue log is obtained, wherein the dialogue log comprises dialogue content and an intermediate result; based on the dialogue content features of the dialogue content, determining evaluation nodes and a plurality of evaluation indexes belonging to the evaluation nodes, the evaluation nodes being used for evaluating sub-processes executed by the large model; based on the dialogue log, performing evaluation analysis on the large model, and determining respective evaluation results of the plurality of evaluation indexes; and based on the respective evaluation results of the plurality of evaluation indexes, determining a target evaluation result for the large model.

The invention relates to the field of large model security testing, and particularly provides a large model output content security testing method and device, and the method comprises the following steps: S1, preparing and managing a test set, a sensitive word library and a regular expression which are required by testing; s2, reading a test set, and obtaining a large model output result according to the test set and the large model interface information; s3, judging whether the output content of the large model is safe or not according to the sensitive lexicon and the regular expression; s4, extracting semantic risk features according to the output content of the large model by using the large model and the oriented Prompt, and automatically storing the semantic risk features after confidence verification; and S5, storing the information result of each request in a file. Compared with the prior art, the test time can be shortened, and the evaluation efficiency can be improved; and the security of the output content of the large model can be effectively evaluated by using a method for dynamically constructing the sensitive word bank by using the output result of the large model.

The invention relates to the technical field of electrical engineering, in particular to a composite material dielectric breakdown multi-scale simulation method and device, and the method comprises the steps: calculating the electric field intensity of a grid based on a preset geometric structure of a nano composite material, and calculating the electric field intensity of the grid based on material parameters and structure parameters of the nano composite material; calculating the molecular interaction energy of the interface between the nanofiller and the matrix; calculating a target breakdown probability of the grid by utilizing the electric field intensity and molecular interaction energy, generating a target state parameter of the grid based on the target breakdown probability, determining a target breakdown strength value meeting a condition by utilizing the target state parameter, and performing Weber distribution fitting processing on the target breakdown strength value to obtain a target breakdown strength value; and determining a simulation result of the breakdown strength of the nano composite material. Therefore, the problems that most of breakdown simulation methods in the prior art are single-scale models, the interfacial effect is not fully considered, errors in the aspect of quantitative prediction of the breakdown strength are large, and the accuracy of breakdown simulation is reduced are solved.

The invention provides an intelligent ore blending method and system based on multi-modal feature fusion. The method comprises the following steps: collecting multi-source data; extracting each modal feature from the multi-source data, and performing weighting processing on each modal feature to obtain a first ore blending scheme; extracting mineral attribute features, semantic features and spatial features of different granularities from the first ore blending scheme by adopting depth separable cavity convolution; fusing the semantic features and the spatial features through a dual-path collaborative enhancement mechanism to generate a candidate ore blending proportion combination; carrying out constraint processing on the candidate ore blending proportion combination, optimizing the candidate ore blending proportion combination by adopting a mixed meta-heuristic optimization algorithm, and outputting an optimal ore blending proportion based on a multi-objective optimization function; and generating a target picture based on the optimal ore blending proportion. And technologies such as a multi-modal feature fusion large model are introduced, so that the reasoning capability is improved, complex ore blending requirements are processed, and the problems of insufficient feature expression and weak dynamic adaptation capability in a traditional ore blending process are solved.

The invention relates to the technical field of large model optimization, and discloses a large model parameter optimization and adaptive adjustment method and device based on reinforcement learning, and the method comprises the steps: obtaining a current operation state of a target large model, and carrying out the comprehensive evaluation of the operation state of the target large model through combining with data features, and obtaining an operation state vector; constructing an intelligent agent, and inputting the operation state vector into the intelligent agent to obtain a parameter optimization strategy; optimizing parameters of the target large model based on a parameter optimization strategy to obtain an optimized large model; calculating a plurality of preset indexes for optimizing the large model by utilizing a reward function to serve as reward results; and adaptively adjusting parameters of the intelligent agent based on the reward result. In training and practical application of the large model, large model parameters are dynamically adjusted by means of an intelligent agent, it is ensured that rapid convergence can be achieved in the initial stage of training, the global optimal solution can be accurately approached in the later stage of training, the training efficiency and quality are effectively improved, and by means of a reward mechanism, the intelligent agent is made to adjust the parameters in a self-adaptive mode, and the generalization ability of the large model is improved.

The invention discloses a low-speed compact wind tunnel experimental device suitable for building outdoor wind environment testing, which comprises a matrix fan system and a modularized segmented structure (sequentially comprising an air inlet section, a rectification section, a model placement area and an air outlet section), and all the sections are connected through flange plates and sealing parts to guarantee the sealing performance. Wherein the matrix fan system is composed of 4 * 4 fans, each fan is subjected to independent frequency modulation through a frequency converter, and the airflow speed is actively adjusted; each section in the modularized segmented structure is an independent module; the air inlet section is used for mixing jet flow generated by the matrix type fan system; a plurality of rectifying parts are arranged in the rectifying section, and the uniformity, turbulence intensity and wind speed gradient of airflow are passively adjusted by configuring different parts, so that specific wind field environment characteristics required by an experiment are accurately simulated; the model placing area is provided with a rotatable disc used for placing a building reduced scale model required by an experiment. The air outlet section is located at the tail end of the wind tunnel experiment device and is provided with an airflow outlet with an adjustable cross section area.

The invention discloses a double-branch multi-scale modeling semantic segmentation method based on a tunnel scene, and the method comprises the following steps: S1, collecting a tunnel face image in a tunnel, constructing a pixel-level annotation data set, and improving the sample diversity through data enhancement; and S2, establishing a context path of the double-branch semantic segmentation model, performing fast down-sampling by using a residual network to extract high-dimensional features, designing a spatial symmetric pyramid module to enhance semantic perception, and finally fusing and outputting average pooling up-sampling and spatial pyramid pooling features. And S3, establishing a spatial path of the double-branch semantic segmentation model, performing down-sampling and pooling processing on the high-dimensional feature map, and adding a dynamic sensing context module to extract a multi-scale feature fusion output result. And S4, carrying out feature fusion on the obtained feature information, and finally predicting a segmented image. According to the invention, the dynamic context sensing module and the space symmetric pyramid module are designed, so that the accuracy of multi-scale classification in a complex scene can be effectively improved.

The invention discloses a wind tunnel-pool combined loading floating fan distributed hybrid test method, which comprises the following steps of: establishing a floating fan multi-body / multi-physical field collaborative reduced scale criterion, dividing a floating fan into two independent physical models and two independent numerical models, and placing the two independent physical models and the two independent numerical models in different laboratories; integrating a physical model, a numerical model, a loading system and a measurement system, and building a floating fan physical-numerical distributed hybrid test platform; and carrying out data interaction and closed-loop control on the physical-numerical hybrid test platform. A fan and a floating platform are adopted as two relatively independent test substructures, test loading or computer simulation can be carried out in different laboratories by establishing different similarity criteria and geometric scale ratios, a collaborative online test and pneumatic-hydrodynamic load loading under the coupling action of wind, waves and flow are carried out through network interaction data, and the test efficiency is improved. The technical problems of environment load high-precision simulation and dynamic coupling response in a reduced scale model test are solved.

The invention discloses an intelligent sweeper path planning method based on complex sweeping road condition sensing and modeling. The invention discloses an intelligent sweeping vehicle path planning method based on complex sweeping road condition sensing and modeling, and belongs to the technical field of intelligent sweeping vehicles. The system comprises a multi-sensor fusion sensing module, a local scene dynamic modeling module and a path planning module. The multi-sensor fusion sensing module obtains dynamic and static target information of road conditions by means of various sensors, the local scene dynamic modeling module constructs a space-time multi-dimensional variable-scale model according to the information, and the path planning module develops an algorithm which comprehensively considers cleaning efficiency and obstacle avoidance safety and plans an optimal driving path. The system can effectively improve the safety and efficiency of intelligent sweeping operation, and is widely applied to the related fields of intelligent sweepers and the like.

The invention provides an automatic bug fixing method based on a large model sensing dynamic program state. The method comprises the following steps of: 1, running verification: verifying existence of vulnerabilities through running, and extracting non-crash constraints; step 2, obtaining an expected state in a repair positioning step, obtaining the expected state based on non-collapse constraint, specifically obtaining the expected state of a collapse position and obtaining the expected state of a potential repair position; 3, a debugging process in the positioning step is repaired, an interface set is provided for the LLM, a program real state is obtained based on program debugging, and static information is obtained based on a tool set; and 4, summarizing debugging information, and generating and verifying patches. According to the method, an actual running state is obtained through program debugging, and an expected state is deduced based on collapse free constraints. By continuously comparing the two states, the nature of the vulnerability can be deeply understood, so that a more accurate and effective repair patch is generated, and the patch repair accuracy, effectiveness and actual availability are remarkably improved.

The invention provides a flow field analysis method capable of optimizing an anti-blocking adjusting stop valve structure, and belongs to the technical field of stop valves, and the method comprises the following steps: S1, based on an energy minimum multi-scale model, dividing each phase medium into a dense phase and a dilute phase through calculation of each phase distribution in each region in a valve; s2, correcting an inter-phase momentum transfer resistance relationship of the liquid-solid two-phase quasi-fluid model, and establishing a mass transfer and drag action relationship between a dilute phase and a dense phase to form a first-stage model; s3, performing medium serialization on the physical image of the first-stage model to form a single dilute phase and a single concentrated phase, establishing a mass exchange and drag action relationship between the dilute phase and the concentrated phase, ensuring that the concentrated phase and the dilute phase meet mass conservation, and constructing a second-stage model; according to the method, the defects of homogeneous hypothesis are corrected through the multi-stage model, the influence of particle aggregation on flow can be reflected, liquid-solid two-phase flow simulation is closer to the actual working condition, and the prediction error of the erosion risk area is greatly reduced.

The invention discloses a dynamic active learning and knowledge graph agent-based airworthiness document review method, which comprises the following steps of: collecting and processing a multi-modal heterogeneous data source to obtain a standardized knowledge base; based on the standardized knowledge base, constructing an aviation domain knowledge system by using dynamic active learning and a man-machine cooperation mechanism; constructing a knowledge graph and performing knowledge enhancement; and dynamically constructing a review context based on a retrieval enhancement generation method in combination with the knowledge graph, and injecting an intelligent agent to perform intelligent data review and report generation. According to the method, automatic review of airworthiness documents is realized through a dynamic active learning-agent review method, and the manual review efficiency is remarkably improved; domain knowledge is injected into an intelligent agent by utilizing an RAG technology, so that the accuracy of a general large language model in an airworthiness review task is improved; through knowledge semantic driving and human-in-the-loop knowledge enhancement, the problem of scarcity of high-quality data in the field is relieved; models of different parameter scales are cooperatively utilized, and computing power resources are saved.

The invention discloses a robot dog target following method based on computer vision. The robot dog target following method comprises the following steps: S10, preparing human body data, human body weight identification data and a single target tracking data set; s20, designing a lightweight human body detection network architecture and a detection human body frame, and performing model training based on the human body data; s30, constructing a lightweight single-target tracking model, selecting a detection frame of a to-be-tracked target as initialization input, receiving an initial frame transmitted by a robot dog camera as a template image, and performing specified template tracking and following; s40, constructing a lightweight full-scale model, starting a relocation mechanism after a lost target is tracked by a single target, and controlling the robot dog to turn around in situ, globally search a specified target and relocate a disappearing target; and S50, based on the tracking result of the single target, converting the target position on the image to an actual three-dimensional coordinate, and controlling the motion displacement of the robot dog to complete the following operation.

A multi-modal large model illusion detection and inhibition method based on attention time sequence difference comprises the following steps: inputting text lexical elements of cue words and visual lexical elements of images into a multi-modal large model, and obtaining an internal attention graph of the decoding stage of the multi-modal large model; then calculating the attention proportion of the visual lexical units at the current generation moment, and making a difference between the attention proportion and the proportion at the previous moment; and if the difference value exceeds a set threshold value, determining that the lexical elements are visual related lexical elements. When the visual related lexical elements are recognized, performing secondary forward propagation of primary visual enhancement to obtain more accurate output; and if not, directly entering the next step of generation. According to the method, visual related lexical elements in text generation are recognized and refined through attention time sequence difference, two-time forward propagation is adopted, the visual attention of second-time forward propagation is enhanced based on a visual attention graph of first-time forward propagation, and illusion can be recognized and corrected on the premise that the language expression ability is not reduced.