119 results about "Model extraction" patented technology

The application discloses a kind of vacuum switch contact ablation intelligent evaluation method based on physical information fusion, belong to vacuum circuit breaker technical field.The method of the present application includes: according to the data set of vacuum switch breaking type obtained from vacuum arc experimental data, construct physical information enhanced support vector machine to predict breaking type of working condition;Vacuum arc simulation model is extracted from each working condition Contact surface energy flow density radial distribution data, and is fitted by Gaussian function;Based on the energy flow density radial distribution curve obtained by fitting, a vacuum arc simulation data set is constructed, and a physical information enhanced Gaussian process regression is constructed to predict the energy flow density of the working condition;Combining the heat transfer model to solve the radial distribution of contact surface temperature, further quantitatively evaluate the ablation degree of contact surface, and output the ablation degree level.The present application embeds the physical law of vacuum arc into machine learning model, realizes the breaking type prediction, energy flow density prediction and contact temperature rise calculation and ablation degree calculation and evaluation.

The application discloses a domain knowledge triple extraction method, system, medium and equipment, and relates to the technical field of knowledge engineering.The domain knowledge triple extraction method and system provided by the application can provide high-quality and efficient data support for triple extraction by analyzing and converting the obtained domain professional text into a structured vector knowledge base which can be efficiently searched, then the domain professional text is subjected to quantitative calculation and fusion screening, the initial entities of domain core high-frequency words which have global high frequency and cross-text universality are mined, the problem of "local optimum and global deviation" caused by traditional single word frequency is avoided, non-core term interference is effectively avoided, and the stability, relevance and efficiency of iterative extraction are improved, and further, the high-frequency word initial entities are subjected to closed-loop iteration through iterative RAG algorithm combined with the structured vector knowledge base, accurate extraction of the large model and iterative correlation entities, the explicit and implicit semantic correlation triples in the text can be mined layer by layer, the strong relevance and structural integrity of the extracted triples are ensured, and the professionalism and accuracy of the model in extracting triples are improved.

A method for spatializing a sound stream including a sound object in a motor vehicle comprises the steps of: extracting (203) the voice using an artificial intelligence model to separate the sound stream into two tracks, a first track including the sound object and a second track including the other sound elements of the sound stream; spatializing (205) the first track so that the sound object is perceived by a user of the motor vehicle as originating from a predefined location in the vehicle; and simultaneously broadcasting (209) the first and second tracks in the vehicle. A spatialization device and a motor vehicle including the device are also described. Figure to be published with the abbreviation: Fig 2

This invention discloses a method, system, device, and storage medium for generating three-dimensional cutting trajectories, comprising the following processes: acquiring a CAD model of the finished product, extracting cutting trajectory lines, and generating a three-dimensional point cloud of the CAD model; scanning the workpiece to be cut to generate a three-dimensional point cloud of the workpiece; registering the three-dimensional point cloud of the CAD model and the three-dimensional point cloud of the workpiece to be cut, calculating the transformation matrix, and performing coordinate system transformation on the CAD model according to the transformation matrix; generating a set of cutting poses based on the cutting trajectory lines and the transformation matrix, performing translation and rotation adjustments, smoothing, sorting, deformation compensation, and robot kinematics checks on the set of cutting poses to generate a set of cutting trajectories containing multiple cutting trajectories; and generating a cutting program based on a recognizable syntax based on the set of cutting trajectories containing multiple cutting trajectories.

The present application relates to the technical field of artificial intelligence large model, specifically to a report generation agent autonomous construction method based on knowledge enhanced large model, which comprises: knowledge extraction on business domain structured documents and unstructured texts and construction of domain knowledge graph, adoption of an improved progressive knowledge injection algorithm to integrate knowledge into general large language model parameters to form a knowledge enhanced base model. Extraction of task planning sequences and operation trajectories generated by historical reports to construct a demonstration sample library, and fine tuning of a primary report generation agent through a behavior cloning strategy. A reinforcement learning training field containing multiple rounds of complex tasks and diversified environmental disturbances is built, interactive data is collected to form a behavior experience pool, decision parameters are iteratively updated based on proximal policy optimization, and autonomous construction of the agent is completed. This method can improve the effect of domain knowledge fusion and enhance the task planning and autonomous decision-making ability of the agent in complex scenarios.

The application discloses a kind of passive domain self-adapting method and system for medical image classification, electronic equipment, comprising: first supervised training model containing encoder and classifier on source domain medical image;Then, the model processes unlabeled target domain image, extracts features and predicts class probability.Utilize prediction probability to optimize Gaussian mixture model, and according to model distribution, target image is classified into class source domain and target specific class.Based on class source domain feature, define class center.New sample is obtained by image enhancement, and the original image is input into the model together, features are extracted and predicted.Combined with prediction probability and feature calculation loss, iterative optimization until convergence.Finally, the model after fine-tuning is used to predict target domain image, and the classification result is output.The application adopts passive domain self-adapting strategy, which avoids the problems of cross-center data privacy and data security.

This invention discloses a rapid interactive and verification method for mold structures based on wireframe preview, applied to a computer-aided design system. The method includes: receiving and loading a parametric 3D structural model of the target mold; parsing the 3D structural model, extracting and registering the driving design parameters, and generating a parameter registry; constructing and displaying a lightweight wireframe preview model representing the structural outline based on the parameter values ​​in the parameter registry; adjusting the design parameters, updating the parameter values ​​in the parameter registry in real time, and driving the wireframe preview model to update synchronously; after confirming that the updated wireframe preview model meets the design intent, driving the computer-aided design system to generate the final 3D solid structure based on the final parameter registry. This invention significantly improves the efficiency of mold structure design by providing real-time preview and rapid adjustment interaction during the design process.

The present application relates to the technical field of engine simulation test, in particular to a thrust characteristic coupling model construction and numerical simulation method about attitude-adjusting engine, comprising: ontology model fusion algorithm platform, structural mechanics analysis calculation platform and multi-dimensional physical field coupling calculation platform. The present application extracts and fuses the models in the digital prototype model library and verification environment model library of the attitude-adjusting engine, couples the corresponding models with the conditional load excitation, checks, and realizes the numerical simulation of the thrust characteristic coupling model under the multi-scene algorithm path matching, adjustment and multi-disciplinary policy algorithm module cooperation. Mainly divided into ontology model fusion algorithm platform, structural mechanics analysis calculation platform and multi-dimensional physical field coupling calculation platform.

The application discloses a natural gas load interval estimation method based on large model coupling working condition clustering, and belongs to the technical field of natural gas pipeline network operation optimization and artificial intelligence load prediction. The method extracts working condition semantic constraints by using a large model, and obtains fuzzy working condition clusters by combining historical operation data clustering; a natural gas load point prediction model is trained for each cluster, and residual probability density distribution is estimated; semantic and numerical weights are fused in real time, and the final point prediction value and the natural gas load prediction interval at the prediction time are obtained by dynamic weighting, which are taken as the natural gas load prediction result and output. The application introduces a large model into the natural gas working condition expression and clustering constraint construction process, and no longer uses the large model as a simple downstream feature generation tool, but solves the problem of interval estimation failure of the natural gas load under fuzzy working conditions such as holiday switching, peak-valley transition and extreme weather through deep coupling of the large model and working condition clustering, so that high-reliability dynamic prediction interval output under complex working conditions is realized.

This invention provides a multi-dimensional user profiling automated segmentation and precise targeting system and method, aiming to solve the problems of single user feature dimensions, lagging updates, and inaccurate matching in existing advertising targeting. The method includes: collecting multimodal data such as user text, behavior, and images; extracting feature vectors from each data source using a model and fusing them into a unified high-dimensional representation; inputting this data into a multi-label neural network to output user interest, behavior, and intent tags; using an improved clustering algorithm to form a strategic audience package; and combining this with ad placement and resource allocation for ad scheduling, while simultaneously updating the user profile and model through a closed-loop user feedback mechanism. This method features innovative aspects such as multi-source heterogeneous feature modeling, minute-level tag updates, and tag-driven clustering and ad delivery, significantly improving audience identification accuracy and ad ROI performance.

The application discloses a real-time detection method for climbing behavior based on space-time information fusion and related components, wherein the method comprises the following steps: first, continuously frame extraction is performed by a camera arranged at a sentry box or an entrance of a park; then, a to-be-detected range is set at a gate entrance or a fence of a video picture; video data in the monitoring video is acquired and continuously frame extraction is performed; each frame of image extracted is sequentially input into an AlphaPose model; human body skeleton information of a single frame of image is extracted; the obtained human body skeleton information is stored in a fixed queue; when the queue length accumulates to a predetermined length, the skeleton information of the queue is input into a 2s-AGCN model for action recognition; and finally, a behavior category and a confidence degree thereof are output. The application can effectively and timely detect whether a human body is making a climbing behavior to ensure the security of the park. The AlphaPose and 2s-AGCN combined technical method is adopted, and the detection precision and speed are improved.

The application provides a music deep forgery detection method and system based on multi-modal fusion, belonging to the technical fields of multi-modal feature fusion and music forgery detection. The method comprises: performing two-way processing on the original music in parallel: the first way extracts aligned musical instruments, vocals and lyrics input through a multi-modal synchronous alignment model, and obtains modal features through a preprocessing model respectively; cross combination is performed on the three modal features to obtain cross-modal features, and then a multi-stage fusion network is used to obtain global fusion features; the second way performs frequency domain analysis on the original complete audio to extract frequency domain compensation features; the two-way features are input into a double-flow fusion network to obtain enhanced global fusion features; and finally classification is performed to obtain music true or false binary classification detection results. The application effectively solves the problems of insufficient use of music deep forgery information by existing methods and difficulty in realizing high-quality AI generated music forgery detection.

The application discloses a kind of based on AI model's place name classification method, system, equipment and storage medium, design surveying and mapping technical field, method includes: convergence multi-source professional surveying and mapping place name data, execute data preprocessing, based on the structured place name text feature vector, generate the semantic vector that can be directly identified by general AI big model, input big model extraction place name deep layer semantic structure feature, while extracting the geographical attribute characteristics corresponding to place name;Structural place name field knowledge system library is built;Introduce the basic reasoning framework built by Transformer big language model and traditional deep network learning technology, adopt knowledge enhancement strategy to optimize model, cross-modal reasoning is carried out by fusing place name semantic structure feature and geographical attribute characteristics, preliminarily determine the category of place name;Adopt prompt engineering driven zero sample and few sample learning mode, generate multiple candidate classification results, output final place name classification result after screening verification.

The present application relates to artificial intelligence technology, disclose a kind of hot video segment identification method, comprising: using pre-trained graphic-text matching model, extract the video feature of each video segment and the text feature of each barrage text, in turn select a barrage text as a text to be matched, the initial video segment corresponding to the text to be matched and the video segment in the preset adjacent range are taken as the video segment set to be matched, the similarity between the text feature of the text to be matched and the video feature corresponding to the video segment set to be matched is calculated, the video segment that meets the preset similarity condition is selected as the matching video segment of the text to be matched, the barrage text after graphic-text matching is classified, according to the number of barrage text in each classification, the hot degree of video segment in the corresponding classification is calculated, and the hot video is determined according to the hot degree.The present application also proposes a kind of hot video segment identification device, equipment and medium.The present application can improve the accuracy of hot video segment identification in medical related video.

This invention discloses an incremental training data filtering method based on historical representation similarity, comprising: inputting a continuously trained incremental data stream and a target model; establishing a historical data set; when the t-th incremental batch arrives, reading the data of that batch as the current batch data set; using the target model to extract the data set to obtain a historical representation set and a current representation set; calculating the representation similarity between samples in the two representation sets, constructing a similarity set, and calculating the similarity statistical score of the current sample; sorting and filtering the data set according to the similarity statistical score to obtain a training data set; using the training data set to update the parameters of the target model and update the historical data set to obtain an updated historical data set. This invention, by constructing a structured filtering mechanism based on historical representation similarity, achieves dynamic selection and scale control of incremental training data, maintaining the stability of model performance while reducing the scale of training data.

This application provides a model migration method, device, storage medium, and program product, particularly relating to the field of artificial intelligence chip technology. The method includes: reading the original model under the original framework; extracting the original definitions of each module and the original configuration data structure of the original model; converting the original definitions of each module into target definitions under the target framework, and simultaneously converting the original configuration data structure into a target configuration data structure under the target framework. Next, based on the model structure definition template of the target framework and the target definitions of each module, the model structure definition under the target framework is automatically generated, achieving automated model structure conversion; finally, based on the model structure definition and target configuration data structure under the target framework, the target model under the target framework is obtained, thus automating model migration and effectively improving the efficiency of model migration.

The present application relates to the technical field of computer model generation, and is a concrete parameter optimization model generation method and system, specifically comprising: establishing a digital twin evolution model fusing prestressed vector field and heat conduction gradient; extracting effective axial force prediction value and interface debonding displacement increment changing with fire duration based on the digital twin evolution model, calculating and obtaining dynamic debonding sensitivity coefficient, synergy factor and topological redundancy response coefficient; generating structure topological robustness maintenance rate; online evaluating deviation tolerance of maintenance rate generated by the current model and target maintenance rate, and adjusting concrete aggregate gradation parameter combination. The present application solves the problem in the prior art that the dynamic collapse risk of the effective contact network of aggregate cannot be characterized, it is difficult to give a gradation optimization boundary with thermal and mechanical robustness, and a technical blind area exists in the safe operation and maintenance of the structure.

This application relates to a method, apparatus, electronic device, and storage medium for recognizing answer information, applied in the field of computer technology. The method includes: acquiring an image to be recognized, the image including a question; acquiring reference answer information corresponding to the question; extracting visual feature information from the image to be recognized based on an end-to-end answer recognition model; recognizing the image to be recognized based on the reference answer information and the visual feature information to obtain answer information corresponding to the question; the visual feature information is used to indicate the spatial positional relationship and contextual information between pixels in the image to be recognized; the end-to-end answer recognition model includes a Conv2former model with an attention mechanism; and the answer information includes handwritten answer information.

The application discloses an AI-based diode model parameter automatic extraction and optimization method, and belongs to the technical field of electronic design.The application solves the problem that the existing diode model parameter extraction method is seriously dependent on manual experience, the process is complicated and time-consuming, and it is difficult to stably obtain high-precision and strong generalization capability model parameters from nonlinear data under the influence of temperature and process deviation.The application constructs an intelligent mapping model from macro electrical characteristics to microscopic model parameters by adaptively denoising and multidimensional feature engineering on measured data, and finally generates a high-fidelity diode electronic design model by combining global optimization of physical constraints, so as to realize high-precision, high-efficiency and full-automatic extraction and optimization of key model parameters such as diode saturation current, ideal factor and series resistance, thereby improving the efficiency, automation level and cross-condition robustness of model extraction.

The application relates to the technical field of code automatic generation, in particular to a system and method for automatic code generation, which comprises the following steps: a requirement analysis module receives user requirements through multi-modal input, and extracts formatted requirement information by using a BERT language model; a template selection module selects a code template from a plug-in template library; a parameter extraction module extracts and verifies key parameters by using a machine learning algorithm; a template filling module fills the parameters into the template and generates a code framework; a framework verification module verifies and corrects the code framework; a code generation module outputs the corrected code framework as a code file and integrates the code file into a development environment of a user project; and a feedback iteration module receives user feedback, analyzes and optimizes system performance. Through full-process automation and intelligent design, the application greatly improves the efficiency and accuracy of code generation, optimizes user operation experience, and can be widely applied to various programming languages and various development scenarios.

This invention discloses a design method for an overvoltage simulation model of composite towers, belonging to the interdisciplinary field of electromagnetic transient simulation and artificial intelligence for transmission line towers. The method includes constructing a multi-wave impedance model and a grounding resistance model and determining the parameters to be optimized; generating training samples composed of lightning current characteristics and tower structural characteristics; constructing a deep neural network surrogate model based on dual-current multi-head self-attention and residual connections, extracting deep representations of the two types of heterogeneous features and capturing their interaction relationships; constructing a near-end policy optimization reinforcement learning framework, and training the surrogate model offline with the goal of minimizing the overvoltage peak deviation; and using the trained surrogate model for adaptive parameter prediction. This invention overcomes the shortcomings of traditional models in representing the frequency-varying characteristics of composite materials, significantly improving the accuracy and efficiency of overvoltage simulation.

Embodiments of the present disclosure provide a model training method, a live face detection method, device, equipment and medium. The model training method comprises: performing self-supervised training on an initial face recognition model by using an unlabeled image dataset, so that the first extractor parameters of the first face recognition model obtained by training have the ability to extract general image data features; performing self-supervised training on the first face recognition model by using a first real live face dataset, so that the second extractor parameters of the second face recognition model obtained by training have the ability to extract face features; performing asymmetric metric learning on the second face recognition model by using a labeled real and fake face mixed dataset, so that the live face recognition model has the ability to judge real and fake faces. The method of the present disclosure significantly improves the ability of the live face recognition model to extract face features and the recognition ability, and can identify fake faces of unknown attack types.

This invention provides an interactive software debugging and automatic repair method based on a large language model, addressing the problems of traditional software debugging relying on manual operation and being time-consuming, as well as the low repair accuracy caused by the lack of a dynamic operating environment in existing static automatic program repair technologies. The method first triggers a program exception by executing a reproduction command and captures the original crash stack information; then, it uses a large model to extract a simplified core function call sequence; subsequently, it constructs a closed-loop debugging mechanism, controlling the large model to generate hypotheses about the root cause of the defect and executing debugger probe commands, iteratively revising the hypotheses based on the dynamic information obtained by the debugger until the defective code is accurately located; finally, it combines the debugging context with the large model to complete root cause analysis and automatically generate and verify executable repair patches. This invention alleviates the "illusion" of large model inference by continuously capturing the underlying dynamic operating state, significantly improving the automation and accuracy of defect repair.

A method and a system for entity disambiguation and forgetting based on a large language model are provided. The method includes: determining an entity disambiguation dataset and a forgetting dataset, constructing a contrastive learning sample, performing data preprocessing; extracting a feature using a LLaMA3 model, adding a projection layer and a contrastive learning module; by constructing a loss function, for each sample, calculating a similarity between its feature representation and positive and negative samples using cosine similarity, measuring an effectiveness of the model in distinguishing positive and negative samples by using contrastive loss, updating model parameters through backpropagation until the model is converged. The present disclosure enhances the discriminative ability of the model based LLaMA3 model architecture and contrastive learning and provides a way to implement a forgetting mechanism. The present disclosure provides new ideas for the research and application of models in the field of natural language processing.

The application discloses a domain adaptation semantic segmentation method based on cross pseudo supervision, and steps include: selecting SYNTHIA and Cityscapes data sets to construct a source domain and a target domain; using real labels to perform double-model cross self-supervised training on the source domain image to obtain a pre-training model; loading the pre-training model in the target domain training set, performing cross pseudo supervision training using pseudo labels, and constructing a cross pseudo supervision double semantic segmentation model; extracting target domain features, designing an attention modulation mechanism, introducing attention modulation loss and entropy consistency loss, and optimizing the model; verifying the model performance on the target domain verification set, and saving the optimal parameters. The application considers the "determinacy" difference of the model, uses attention modulation and entropy consistency loss, improves the generalization ability and robustness of the model, and adapts to semantic segmentation tasks in different fields or environments.

The application provides a substation equipment defect identification method and device and electronic equipment, and relates to the field of image recognition. In the method, an infrared image, an electric field leakage map and a visible light image are obtained by a multi-channel imaging system deployed on the substation site, a multi-channel image tensor is generated and input into a multi-channel identification model to extract and fuse features. The fused features are input into a YOLOv8 backbone network, a joint attention field is constructed in combination with a device prior structure, high-confidence candidate boxes are generated, and a non-maximum suppression is performed to obtain detection results. A frame residual tensor is constructed for the detection results to perform time series modeling, thereby improving the detection effect. For a device with complex occlusion, the structure contour is completed through an edge prediction path, and finally the target boundary and defect positioning information are output. The technical solution provided by the application facilitates defect identification of substation equipment.