21 results about "Interactive modeling" patented technology

The invention provides a task unloading decision-making method based on multi-agent reinforcement learning. The method comprises an execution stage and a training stage. In the training stage, the centralized critic network can obtain global information of all agents so as to guide optimization of strategies of all the agents. In the execution stage, each agent only needs to realize the distributed unloading decision of each wireless device through a local actor network according to the local observation of the agent. According to the method, the global state and the joint action are evaluated by adopting the centralized value evaluation network, and the resource competition and coupling relationship between the intelligent agents is explicitly learned through the interactive modeling structure, so that the value estimation deviation caused by a simple splicing structure is reduced, the joint value evaluation precision is improved, and a more accurate updating direction is provided for each actor.

The invention relates to a VR-based multi-scene learning state modeling and learning effect evaluation method and system, and the method comprises the steps: defining a student learning state as a dynamic interaction modeling object and a static modeling object according to the VR modeling requirements and the demands of multi-scene learning effect evaluation for modeling, and carrying out the decomposition; drawing a multi-scene learning effect evaluation modeling network diagram; a teaching scene modeling topological structure is extracted from the network diagram, multi-scene learning state modeling parameter quantification is carried out, and a student multi-scene learning state model is constructed through a virtual reality modeling technology; and based on a multi-scene learning target, constructing a learning effect comprehensive evaluation parameter, carrying out multi-scene learning effect comprehensive evaluation and each scene process learning effect evaluation, automatically generating an evaluation report, and synchronizing the evaluation report to a teacher end and student multi-scene learning state model. According to the method, scene coverage is more comprehensive, data association is tighter, evaluation precision is improved, unified modeling of multi-scene data is achieved, and the visualization effect is better.

The invention discloses a Web service causal depolarization recommendation method and a Web service causal depolarization recommendation system based on attention enhancement contrast learning. The method comprises the following steps: 1) randomly sampling a negative sample for each positive sample in user-project interaction, and dividing into two mutually exclusive training sets, namely, a more popular positive sample set and a more popular positive sample set, according to project popularity; 2) based on the two mutually exclusive training sets, executing a graph convolutional network and an Attention attention mechanism, and generating an interest representation matrix and a consistency representation matrix; 3) performing regularization constraint on the embedding space of the user and the project by introducing a triple loss function on the basis of a comparative learning representation distribution optimization framework; 4) performing joint optimization on a plurality of losses of interest modeling, conformity modeling, interactive modeling, independence constraint and comparative learning, and optimizing the model through a curriculum-type training strategy for gradually tightening constraint boundaries, and 5) calculating a comprehensive preference score based on the final interest representation matrix and the consistency representation matrix, and outputting a debiased recommendation result.

The invention is applied to the technical field of new energy power generation prediction, and discloses a wind power prediction method based on non-overlapping multi-scale division and interactive modeling, and the method comprises the specific steps: firstly, carrying out the space-time embedding of historical wind power data, and obtaining the space information; secondly, a multi-scale token length set is set, non-overlapping multi-scale partitioning is carried out on the embedded sequence, and tokens of different scales are generated so as to reduce redundancy caused by overlapping; then, constructing a hierarchical interactive encoder, extracting features layer by layer from a small scale to a large scale, and embedding the small-scale potential representation of the previous layer into the input of the current layer in the large-scale encoding process to realize the guidance of the large scale by the small-scale features; and finally, fusing and mapping each scale feature output to obtain a future wind power prediction value, and training and optimizing the model by adopting mean square error loss. Verification on actual wind power plant data shows that the method can effectively improve the accuracy of wind power prediction.

This application relates to the interdisciplinary field of artificial intelligence and reliability systems engineering, and provides a method and apparatus for intelligent generation of equipment fault behavior logic models based on LLM (Limited Linear Modeling). This method combines a vector database to achieve enhanced retrieval, constructs an agent-based framework for generating equipment fault behavior logic models, and then generates structural elements in the agent for each business component involved in the equipment fault behavior logic. Based on the model structure, the results are integrated and merged, ultimately achieving a complete and runnable equipment fault behavior logic simulation model. This method integrates the interactive modeling capabilities of agents with the semantic understanding and logical reasoning capabilities of large language models, enabling intelligent generation of fault behavior models for complex equipment and effectively improving the generation efficiency of equipment fault behavior models.

The invention discloses a method and a system for enhancing multi-scale and multi-modal intelligent interactive modeling, and a medium. The method comprises the following steps: generating vertexes and 3D human body grids by using an SMPL model; performing feature extraction on the sketch by using a double-branch encoder, wherein the double-branch encoder comprises a segmentation encoder of a pre-trained human body sketch and a sketch local feature encoder of a sketch based on ResNet34; enhancing the ability of the model to capture a key region in the feature map by using a spatial feature enhancement module; using a multi-scale context fusion module to extract a multi-scale feature map from an intermediate layer of the double-branch encoder; and the output is converted into a 3D human body grid through a multi-branch regression device and an SMPL model. According to the invention, key spatial details are reserved through the multi-scale context fusion module, so that the model can focus key information on feature maps of different levels and scales. According to the invention, higher precision can be realized in 3D human body reconstruction, and optimal performance can be achieved.

The invention discloses a common preference enhanced collaborative recommendation method and device, and the method comprises the steps: obtaining user implicit interaction data in a target domain and a source domain, and constructing corresponding user features and commodity feature representations; carrying out fusion coding on the target domain user features, the source domain user features and the target domain commodity features, and generating cross-domain migration features facing target domain recommendation tasks; performing fusion iteration de-noising processing on the cross-domain migration features, gradually inhibiting source domain noise interference and strengthening migratable collaborative preference information through multiple rounds of feature fusion and weight screening operation, and generating de-noised cross-domain user feature representation; after denoising is completed, user structure constraint is introduced, and consistency constraint is carried out on the similarity relation between the target domain users before and after feature enhancement, so that the overall distribution stability of the target domain user interest structure is kept; and finally, performing interactive modeling on the final user representation and the target domain commodity features to obtain a matching relationship between the user and the commodity. The device comprises a processor and a memory. Through a collaborative modeling mode of fusing iterative denoising and structural constraint, the problems of noise interference in cross-domain recommendation and user similarity structure change before and after feature migration can be relieved, and the stability and accuracy of a recommendation result are improved.

The invention relates to the technical field of trajectory prediction and planning, in particular to a parallel iterative joint prediction planning method considering bidirectional interaction characteristics, and the method comprises the steps: constructing a bidirectional interaction modeling mechanism considering future trajectory interaction; iteratively executing interactive trajectory prediction and interactive trajectory planning based on a bidirectional interactive modeling mechanism in parallel until a preset condition is met, so as to respectively obtain a final planning code and a final prediction code; and decoding the final planning code and the final prediction code to obtain a multi-modal prediction trajectory and a planning trajectory of a planning time domain, and obtaining a final prediction trajectory and a planning trajectory according to the confidence coefficient. Therefore, the problems that the coupling relation between prediction and planning is difficult to fully reflect and the vehicle track cannot be timely adjusted according to the traffic environment change due to the fact that only the one-way influence of prediction on planning is established in the related technology are solved; therefore, the problems of conservative planning, influence on road traffic efficiency, even traffic accidents and the like easily occur in a complex dynamic scene.

The invention discloses a technology debt elimination opportunity prediction method and system based on multi-modal data fusion. The method comprises the following steps: firstly, extracting technical debt features including numerical features, text semantic features and files related to Git submission records, and constructing a file covariant graph based on the submission records; secondly, performing multi-modal feature coding on the numerical features, the text semantic features and the graph structure features based on the file covariant graph; then, based on a cross-modal attention mechanism, interactive modeling is carried out on the coded numerical value features, text semantic features and graph structure features, and three paths of interactive representations are generated respectively and spliced into a joint representation; and finally, inputting the joint representation into a multi-layer perceptron regression network, and predicting a numerical value of the residual life of the technology debt through layer-by-layer calculation. According to the method, static code measurement, text semantic information and graph structure features can be fused, key factors are dynamically weighted through attention, and by means of a deep classification model, high-precision prediction of the technical debt remaining life is achieved.

The invention discloses an interactive perception trajectory prediction method for heterogeneous agents in a shared space, and the method comprises the steps: S1, encoder-historical trajectory and interactive feature extraction, applying Transform-based spatial modeling at each time step, processing historical trajectory data, and extracting and fusing the motion features of each agent and the spatial interactive features of a historical period; and S2, decoder-future trajectory iterative prediction and interactive modeling: iteratively predicting a future trajectory based on an encoder output state, and dynamically and sparsely updating interaction between the intelligent agents in the prediction process to simulate future interaction evolution. According to the method, heterogeneous characteristics can be accurately perceived, future interaction can be foreseen, and the calculation efficiency is high.

This application relates to the interdisciplinary field of artificial intelligence and reliability systems engineering, and provides a method and apparatus for intelligent generation of equipment maintenance and operation support models based on LLM (Limited Linear Modeling). This method combines a vector database to achieve enhanced retrieval and constructs an agent-based framework for generating equipment maintenance and operation support models. Then, for each business component involved in the equipment maintenance and operation support model, structural elements in the agent are generated one by one. Based on the model structure, the results are integrated and merged, ultimately achieving a complete and operational equipment maintenance and operation support model. This method integrates the interactive modeling capabilities of agents with the semantic understanding and logical reasoning capabilities of large language models, realizing the intelligent generation of maintenance and operation support models for complex equipment, and effectively improving the generation efficiency of equipment maintenance and support models.

This invention relates to a method for force feedback modeling of intravenous puncture in virtual training for intravenous therapy nurses. The method comprises two parts: hierarchical interactive object construction and force feedback mapping, and hierarchical tactile model establishment. By performing hierarchical interactive modeling of the arm and blood vessel models, a positional mapping relationship between the force feedback device and the virtual needle is established, and the contact state is dynamically updated through interaction between the needle tip collider and the arm and blood vessel layers. Damped tactile feedback models and puncture-feeling tactile feedback models are established for the arm and blood vessel layers respectively, and force feedback signals are calculated and output in real time at a preset refresh frequency. Simultaneously, stiffness, damping, static friction, and dynamic friction parameters are visualized and adjusted, and stress feedback parameters are configured according to different body types. This method can improve the realism and adaptability of force feedback in virtual training for intravenous puncture.

The invention relates to the technical field of artificial intelligence and reliability system engineering crossing, and provides an LLM-based equipment fault behavior logic model intelligent generation method and device. The method combines a vector database to realize enhanced retrieval, constructs an equipment fault behavior logic model generation framework based on an intelligent agent, then generates structural elements in the intelligent agent one by one for business components involved in equipment fault behavior logic, and completes integration and combination of results based on a model structure. And finally, a complete and operable equipment fault behavior logic simulation model is realized. According to the method, the interactive modeling ability of the intelligent agent and the semantic understanding and logical reasoning ability of the large language model are fused, intelligent generation of the fault behavior model of the complex equipment is realized, and the generation efficiency of the equipment fault behavior model is effectively improved.

The invention relates to the technical field of geological exploration, and discloses a mine modeling method, device, equipment and medium, and the method comprises the steps: obtaining the topographic data of a goaf; constructing a fault model by using an interpolation method based on the topographic data; based on topographic data, constructing a thin-layer ore bar model by utilizing spectral characteristics of thin-layer ore bars; and fusing the fault model and the thin-layer ore bar model, and determining a mine model corresponding to the goaf. According to the scheme, the terrain data and the interpolation method are utilized, the fault model is constructed, the fault constraint surface is formed, and the thin-layer ore bar model is constructed by utilizing the spectral characteristics of the thin-layer ore bars to enhance the boundary recognition precision of the thin-layer ore bars, so that the fault space distribution and thin-layer ore bar distribution rules are recognized on the basis, traditional manual interactive modeling is replaced, and the recognition precision of the boundary of the thin-layer ore bars is improved. And constructing a self-adaptive three-dimensional grid model.

The invention relates to the field of new energy centralized control, and discloses a new energy centralized control scene-oriented AI model cloud edge collaborative management system, and the system comprises a model development unit which is used for selecting at least one modeling mode selected from automatic modeling, visual modeling and interactive modeling in combination with the business scene demands of new energy centralized control to construct an AI model; the model training and optimizing unit is used for training, verifying and evaluating the AI model output by the model developing unit in sequence; the AI asset management unit is used for performing version management on the trained AI model, and performing layered packaging on the AI model according to an application layer, a model layer, a configuration layer and an environment dependence layer according to a cloud edge collaborative deployment requirement; and the cloud edge collaborative management unit is used for issuing and deploying the target model package to a specified area side or station side edge node. According to the invention, by constructing a hierarchical progressive intelligent decision-making process, automation and optimization of modeling mode selection are realized.

The invention discloses a multi-view three-dimensional reconstruction method and system based on shape prior and corresponding structure learning, and belongs to the field of computer vision. Comprising the following steps: extracting global semantic information of a multi-view image by adopting a pre-trained attention model, and constructing shape priori of a target object; performing interactive modeling on each view angle feature by using a corresponding structure learning network, firstly obtaining an elementary corresponding structure, then introducing shape prior to generate an enhanced local feature, obtaining a robust corresponding structure through feature alignment, and finally performing multi-view feature aggregation; a refined three-dimensional reconstruction network is adopted, and based on aggregated features, voxel representation with fine-grained geometric details is generated through alternate processing of a multi-layer up-sampling and memory maintenance module. The invention aims to solve the problem that the reconstruction result is incomplete due to self-shielding in the prior art. By introducing shape prior to guide feature learning and aggregation, the robustness, integrity and detail precision of three-dimensional reconstruction under the shielding condition are remarkably improved.

This invention discloses an interactive design method for personalized customization solutions oriented towards changing requirements. The method begins with extracting user interactive requirement features, and then uses a knowledge graph to verify the support of new requirements against existing manufacturing capabilities and embed knowledge, constructing a two-way verification mechanism for requirements and capabilities. Based on this, a semantic plasticity valve is built, incorporating a cross-attention mechanism, a multilayer perceptron, and a relevance perception gating unit. This allows for in-depth interactive modeling of sub-intents at different interaction rounds, integrating dynamically changing historical and new requirements into a unified fusion requirement matrix. Finally, a customized solution is generated through precise matching with nodes within the knowledge graph. This method uses the existing manufacturing capabilities of the production workshop as a hard evaluation criterion to verify their support for new and changing requirements, thereby achieving interactive design of customized solutions based on the effective integration of historical and incremental requirements.

The invention provides a machine learning assembly line framework oriented to end-side cloud collaboration and an implementation method. The framework is oriented to an end-side cloud collaboration architecture and a machine learning algorithm model for processing complex problems. A machine learning modeling engine oriented to workflow, componentization modeling and assembly line modeling is constructed, visual interactive modeling design is supported, algorithm component libraries such as big data processing, machine learning and deep learning and user-defined component functions are provided, and services such as hyper-parameter search, timed task scheduling, process visualization and online tool development are supported. The full-life-cycle efficient service support of a machine learning algorithm from code development to model construction to operation scheduling to issuing deployment is realized, and high-frequency modeling, training, experiment and deployment are supported, so that the requirements of frequent iteration and rapid application of an algorithm model in an actual task of a user are met.

This invention discloses a robot control method and system based on a multimodal thought chain, belonging to the field of robotics technology. The method employs the mCoT-VLA framework, achieving deep interactive modeling of robot visual observation, language commands, and future state prediction through an integrated design of multimodal thought chain reasoning, hybrid attention mechanisms, and cross-modal semantic alignment. This significantly improves the robot's task reasoning ability, execution success rate, and generalization robustness in complex dynamic environments. The multimodal thought chain reasoning adopts a decomposed reasoning architecture, adding an explicit multimodal intermediate reasoning stage before action generation to break the end-to-end model of direct mapping between perception and action in traditional VLA models. This invention can effectively improve the task reasoning ability, execution success rate, and generalization robustness of embodied agents in complex dynamic environments.

The invention relates to the technical field of artificial intelligence and reliability system engineering crossing, and provides an LLM-based equipment maintenance usage guarantee model intelligent generation method and device. According to the method, enhanced retrieval is realized in combination with a vector database, and an intelligent agent-based equipment maintenance and use guarantee model generation framework is constructed. And then, aiming at business components involved in the equipment maintenance and use guarantee model, generating structural elements in the intelligent agent one by one, and completing integration and combination of results based on a model structure, thereby finally realizing the complete and operable equipment maintenance and use guarantee model. According to the method, the interactive modeling ability of the intelligent agent and the semantic understanding and logical reasoning ability of the large language model are fused, intelligent generation of the maintenance guarantee and use guarantee model of the complex equipment is realized, and the generation efficiency of the equipment maintenance and guarantee model is effectively improved.

The embodiment of the application provides a digital human interaction system and method based on multi-modal emotion recognition, and belongs to the technical field of digital human interaction; the system comprises a multi-modal perception module, which is used for collecting multi-modal data, pre-processing the multi-modal data, and generating standardized data flow; a cross-modal fusion and emotion recognition module, which is used for interactive modeling of multi-modal features, and outputs current emotion labels and emotion intensity; a reaction planning module, which is used for generating a composite reaction strategy; and a digital human rendering module, which is used for mapping the composite reaction strategy into control signals corresponding to voice, facial expression and action respectively, and driving the digital human to perform corresponding voice output, facial expression change and limb action through the control signals to realize interaction. Through cross-modal graph neural network and contrast learning, the multi-modal data is deeply fused, the weight is dynamically adjusted in combination with the modal confidence, and the emotion recognition accuracy and robustness are improved.