525 results about "Model architecture" patented technology

The invention relates to the technical field of fault diagnosis, in particular to a wind driven generator fault diagnosis method and system based on Mamba-ResNet. The method comprises the following steps: carrying out feature extraction and feature fusion by utilizing preprocessed data, namely constructing adaptive window short-time Fourier transform (AW-STFT) to carry out dynamic time-frequency resolution analysis, carrying out parallel feature extraction and constructing a multi-dimensional heterogeneous feature vector, and carrying out a cross-modal adaptive gating fusion mechanism based on a bidirectional cross gating unit; the method comprises the following steps: constructing a Mamba-ResNet hybrid deep network model architecture; performing model training on the constructed network model architecture; and performing fault diagnosis on the wind driven generator by using the trained model architecture. A tedious manual feature design process in a traditional method is avoided, and the automation level and adaptability of a diagnosis system are remarkably improved.

The invention provides a medical image segmentation method based on an AFMHiFormer. The method comprises the steps that firstly, a multiple data enhancement module is provided, and the data distribution diversity is improved while the enhancement stability is guaranteed; secondly, a segmentation model AFHiMFormer is constructed, and the model architecture adopts a double-branch encoder and a multi-scale decoder; thirdly, a feature enhancement module is provided to construct a dynamic complementation mechanism of semantic enhancement and boundary modeling; fourthly, a multi-scale feature fusion module is introduced, multi-scale context information is captured through parallel hole convolution with different expansion rates, and self-adaptive fusion of global and local features is achieved; and fifth, a cross-scale fusion module is designed in the multi-scale decoder, so that the deep layer branch and the shallow layer branch are efficiently fused in a multi-level feature space. According to the method, the advantages of CNN and Transform are combined, dynamic fusion of local and global features is realized by providing a new module, and a remarkable performance advantage is shown in a medical image segmentation task.

The invention discloses a reinforcement learning training method and system for relieving illusion of a multi-modal large model, and belongs to the field of reinforcement learning training of a multi-modal large language model. Firstly, a planning and visual description generation step is introduced in an early stage to guide a model to perform structured reasoning, then a grouping relative strategy optimization algorithm is used, reward values are calculated for multiple candidate responses generated by the model after cold start, and particularly, a visual perception reward mechanism is set. The reward mechanism evaluates the consistency of the generated text description and the visual information by using an external large language model. Then, based on a vision description attention score advantage distribution method, learning of the model on key vision signals is dynamically enhanced, and the perception ability of the model on the vision signals is improved; and finally, the perception and reasoning performance of the model is further improved by adopting multiple rounds of rejection sampling and supervised fine tuning. The scheme does not depend on a model architecture, the extra overhead is small, the illusion problem caused by early image-text inconsistency is effectively solved, and the accuracy and the reliability are improved.

Predicting a future state, such as a future position and / or orientation (i.e., pose), of an object may comprise classifying, by a first machine-learned model, a lane the object may occupy and classifying, by a second machine-learned model, a target pose the object may occupy. A third machine-learned model may determine an offset from the target pose that may be used to determine a predicted (future) pose of the object by applying the offset to the target pose.

The invention relates to a wind power cluster short-term power prediction method and device of a space-time diagram neural network fused with physical information and computer equipment, and the method comprises the steps: obtaining related information data of each wind power plant in a wind power cluster, and carrying out the preprocessing; forming a physical prior data set through an engineering analysis model fusing the wake flow analysis model and the blocking effect model; taking each wind power plant as a node of the graph, constructing graph structure data for predicting the power of the wind power plant, and forming a dynamic adjacent matrix; constructing a space-time diagram neural network WB-STGNN model architecture comprising a diagram convolutional neural network module, a gating time convolutional network and a multi-layer perceptron; the method comprises the following steps: pre-training by using a physical prior data set, and then performing formal training based on historical power data and a dynamic adjacency matrix to obtain a space-time diagram neural network WB-STGNN model; inputting the wind speed of the prediction day, and predicting the active power of the whole wind power cluster in 24 hours of the prediction day. By adopting the method, the precision and efficiency of wind power cluster power prediction can be effectively improved.

Systems and methods for an artificial intelligence model architecture that involves a first artificial intelligence model trained to map a plurality of entities to ranked documentation from a documentation source, and a second artificial intelligence model that comprises a language model trained to generate an additional query to run on the plurality of documents from the documentation source. By training the second model to generate additional queries as entities and / or links are discovered, the system may quickly and efficiently determine links and / or potential resolutions as well as received feedback thereon.

The invention provides a model generation method and device, a storage medium and a program product, relates to the technical field of computers, and solves the problem that a large language model is relatively large in illusion. The method comprises the following steps: acquiring a first output and a second output of each attention head in a plurality of attention heads in an initial model for a first input; for the multiple attention heads, based on the first output of the attention heads and the second output of the attention heads, the sensitivity of the attention heads to the context in the first input is determined, and the sensitivity of the multiple attention heads to the context in the first input is obtained; determining an enhanced attention head from the plurality of attention heads based on the respective sensitivity of the plurality of attention heads to the context in the first input; and in the model architecture of the initial model, adjusting the weight of the attention enhancing head based on the sensitivity of the attention enhancing head to the context in the first input, and obtaining a target model.

Various aspects of the present disclosure relate to methods, apparatuses, and systems that support performance monitoring of a two-sided model. For instance, implementations provide a two-sided model architecture composed of a user equipment (UE) component and a network entity component. Accordingly, the present disclosure supports performance monitoring of such two-sided models such as to determine whether models are accurately characterizing channel state information (CSI)-related data. For instance, model performance can be monitored at a UE, at a network entity (e.g., gNB), and / or at both a UE and a network entity. When performance of a model is determined to be outside of specified parameters, the model can be updated and / or replaced.

The invention relates to a sparse view angle pose-free scene reconstruction method and system based on 3DGS, belongs to the field of computer vision and three-dimensional reconstruction, and solves the problems of easy failure and poor geometric consistency in a sparse view angle or weak texture environment due to dependence on accurate camera pose priori. Constructing a double-flow sensing module containing semantic flow and geometric flow, extracting semantic features by using a visual basic model, and extracting an explicit geometric corresponding relation by using a dense feature matching network, so as to regress relative camera pose under pose-free priori; a geometric guidance depth refinement module combining a potential diffusion model architecture and Pluecker ray coding is introduced, and scale fuzziness of monocular depth estimation is eliminated through a depth residual prediction mechanism; and based on the micronizable Gaussian rasterization, performing end-to-end optimization by using a self-supervised loss function including rendering consistency, reprojection and epipolar geometric constraint. According to the method, high-fidelity three-dimensional reconstruction is realized without supervision of external parameter true values, and geometric stability and rendering quality in a complex scene are improved.

According to the converter valve temperature field reconstruction method and device based on the reduced-order model provided by the invention, the model precision and the response speed are improved by constructing a reduced-order modeling architecture in which the primary function dynamic optimization network and the double-layer coefficient prediction network cooperate. Specifically, working condition parameters are input into a pre-trained primary function dynamic optimization network to obtain a primary function correction, and an initial primary function is corrected according to the primary function correction, so that the primary function can adapt to different working condition changes, and the problem of prediction precision attenuation caused by primary function mismatch is effectively solved. Furthermore, a target modal coefficient is determined through a double-layer coefficient prediction network combining a global layer and a local layer, and local nonlinear features are finely compensated while the macroscopic trend is captured. And finally, performing linear superposition on the target basis function and the target modal coefficient, and reconstructing to obtain the temperature field of the converter valve. Therefore, on the premise that details of the high-precision three-dimensional field are reserved, the reconstruction speed of the temperature field is increased, and the reconstruction precision is considered at the same time.

The invention provides a space-air-ground integrated ecological monitoring method and system based on multi-source data fusion, and the method comprises the steps: collecting a multi-source heterogeneous data set, carrying out the space-time alignment, multi-scale fusion and dynamic deduction, and generating an ecological change dynamic deduction result; and carrying out ecological monitoring early warning and decision support based on the result. According to the method, cross-platform data is safely cleaned and enhanced through a federated learning framework, and data modal differences are eliminated; deep fusion static feature regression and time sequence trend prediction are carried out by using a double-model architecture, and microenvironment detail changes and long-term ecological trends are synchronously captured; and finally, through a decision strategy model driven by a double-target reward function, quantifying an ecological restoration cost-benefit ratio and a stability gain as an optimal path, forming a'data fusion-dynamic deduction-decision support 'full-link closed loop, improving ecological monitoring space-time continuity, trend prediction accuracy and management decision scientificity, and improving ecological monitoring efficiency. The contradiction that in the prior art, a user can see widely but cannot see finely, and the user can measure finely but does not tend to measure is overcome.

The invention relates to the technical field of biped robot reinforcement, and discloses a historical action sequence-based biped robot reinforcement learning control method and system. The biped robot reinforcement learning control method is applied to reinforcement learning control equipment and specifically comprises the following steps that S101, a floating base frame is adopted to load a biped robot URDF model in an Isaac Gym simulation platform, ten active joint freedom degrees are configured, and a control interface is constructed to output a 10-dimensional motor position command vector; the motion robustness and adaptability of the biped robot in a dynamic environment are remarkably improved by constructing a Transform time sequence modeling architecture fusing a historical state-action sequence, and the historical observation and action sequence is combined with sine-cosine position coding to be input into a multi-layer Transform encoder, so that the motion robustness and adaptability of the biped robot in the dynamic environment are improved. The self-attention mechanism is used for capturing the long time sequence dependence relation of foot-ground contact force change, topographic relief and the like, and the problem of gait instability caused by limited local observation in a traditional method is solved.

The invention discloses an oil and gas field integrated dynamic simulation construction method and system based on artificial intelligence, and the method comprises the steps: S1, carrying out the space-time alignment and standardized cleaning of a multi-source heterogeneous data signal, and generating a fusion data signal with unified space-time; s2, a physical information embedded deep learning model architecture is constructed, the architecture receives the fusion data signal, and a mixed training target signal fusing data driving and mechanism driving is generated; s3, training the deep learning model by using a fusion data signal of a historical time sequence, and finally obtaining a convergent intelligent agent model; and S4, feeding the current oil and gas reservoir state parameters and the production control parameters into the intelligent agent model as input signals, and directly outputting a dynamic prediction signal. The oil and gas field integrated dynamic simulation construction method and system based on artificial intelligence can solve the problem that traditional oil and gas field numerical simulation calculation is low in efficiency.

The invention discloses a box-type substation transformer fault diagnosis method and system based on artificial intelligence, relates to the technical field of power equipment fault diagnosis, and solves the problems that a sensor mechanism lacks a dynamic redundancy and verification mechanism, the feature contribution degree difference is not considered in cloud feature fusion, and the fault diagnosis efficiency is low. The characteristic quality is difficult to guarantee; and the diagnosis model architecture is single. Comprising the following steps of firstly constructing a sensor cluster network, dividing a main unit, a standby unit and a verification unit through a multi-index dynamic weighting algorithm, performing cooperative work, then generating a lightweight feature set through edge node preprocessing, constructing a normalized multi-dimensional feature matrix through cloud data integration, and finally constructing an edge-cloud secondary diagnosis architecture. The edge quickly screens the fault through a random forest, and the cloud end dynamically adjusts the feature fusion proportion by adopting a double-branch fusion model in combination with a gating circulation unit, and outputs the fault type and confidence.

The invention provides a heterogeneous model knowledge migration method based on federated learning. The method comprises the following steps: a data collaborative preparation stage; screening local data, performing multi-dimensional labeling, and performing cooperative training and federal modeling preparation on a plurality of participants; a privacy initialization stage; privacy protection processing is carried out on local data, and local model parameters are initialized; a self-learning stage; each participant completes localized updating of the model parameters by using the local data set; a mutual learning stage; the server extracts valuable knowledge fragments from the knowledge representations uploaded by the participants through the super network for mixing, and updates model parameters of the participants; a server-side super network module is updated and iteratively cycled; and the server side aggregates the valuable knowledge fragments, updates the super network parameters and then returns to the step S3, and loop iteration is carried out until the model converges. The problems that computing resources of participants are limited, generalization is insufficient due to model heterogeneity, and a knowledge migration technology cannot adapt to a heterogeneous model architecture are solved.

The invention relates to the technical field of urban inland inundation rapid prediction, and discloses an urban inland inundation rapid prediction method based on deep learning, and the method comprises the following steps: S1, collecting historical meteorological data, landform data, urban drainage system data and historical inland inundation event data; s2, performing data cleaning on the collected data, removing noise, filling missing values, and processing abnormal values; s3, constructing a deep learning model architecture; when urban inland inundation risk prediction is carried out, multi-source heterogeneous data are integrated and standardized, and a unified spatial-temporal characteristic analysis framework is constructed, so that the system can eliminate magnitude differences of weather, terrain and drainage system data, and data comparability of different regions is ensured; and meanwhile, dynamic feature extraction is performed on real-time rainfall data by using a deep learning model, an abnormal fluctuation rule of meteorological elements is identified, the waterlogging risk pre-judgment capability of extreme weather events is improved, and the stability and credibility of a prediction result are enhanced.

The invention discloses a system for realizing bidirectional conversion between a model and a code, and the system comprises a meta-model architecture module which is used for constructing a unified model specification; the meta-model engine module is used for generating, analyzing and processing the meta-model architecture module; and the bidirectional conversion module is used for realizing forward and reverse synchronous conversion between a zero code model and a Java model serving as a program code by adopting the meta-model engine module. By means of the scheme, a unified specification with the meta-model architecture as the core is established, a powerful meta-model engine is used for automatic processing, and the barrier between a zero-code visualization model used by business personnel and bottom-layer Java codes written by developers is seamlessly broken through. The synchronization capability of the model and codes ensures the continuous alignment of business design and technical implementation, and effectively supports the quick iteration of agile development and business.

The invention provides a crop growth monitoring analysis system and method based on machine learning, and belongs to the technical field of crop growth monitoring. The monitoring analysis system comprises a data acquisition module used for acquiring biomechanical characteristic data of crops and energy exchange data of an SPAC system; the data processing module is used for preprocessing the data to generate a mechanical feature set and an SPAC feature set; the model construction module is used for constructing a multi-layer model architecture which comprises feature association, crop growth state evaluation and crop growth abnormal condition cause traceability reasoning; the model training module is used for constructing a training data set, training the multi-layer model architecture in combination with a transfer learning strategy, and updating the multi-layer model architecture by utilizing an optimization mechanism; and the monitoring analysis output module is used for visually displaying a crop growth monitoring analysis result and generating corresponding farming suggestions and measures. According to the method, the accuracy of stress risk early warning is improved, the model adaptability is improved, and the long-term monitoring precision stability of the model is improved.

The invention provides a construction method of an ocean observation and exploration large model, and belongs to the technical field of large models.Multi-mode original data are collected by constructing a multi-source ocean data collection matrix, an environment change degree vector is established, preprocessing and noise reduction are conducted on the original data by adopting a nonlinear matrix mapping algorithm based on a Gaussian kernel function, and the large model is constructed. An ocean observation and exploration large model architecture of a liquid neural network structure is constructed, different branches are made to process input data of different dimensions by means of asymmetric design, a super sparse reconstruction matrix is established, and high-dimensional original data are reconstructed from low-dimensional observation by means of a compressed sensing reconstruction mechanism. And finally, a supervised training process is executed to optimize model parameters so as to complete the construction of an ocean observation and exploration large model, and the technical problem that high-precision fusion modeling of ocean multi-modal observation data is difficult to realize under the conditions of spatial-temporal distribution sparsity and data isomerism is solved.

The invention discloses a digital circuit design method and device, equipment and a medium, and relates to the technical field of digital circuits, and the method comprises the steps: determining a target type of a to-be-designed digital circuit, and collecting a first training set corresponding to the target type; each training sample in the first training set comprises an input feature item used for designing a target type of historical digital circuit and the actual result quality of the historical digital circuit, and the input feature item comprises a high-level language source code and an HLS constraint parameter; constructing an initial prediction model based on the multi-layer perceptron mixed structure, training the initial prediction model by using the first training set, and searching and optimizing the model architecture based on the network structure to obtain a target prediction model; and predicting the actual result quality corresponding to each parameter combination in a target design space of a to-be-designed digital circuit by using the model, and designing the digital circuit by using the parameter combination of which the quality meets the Pareto frontier. And predicting an accurate actual quality result by utilizing a prediction model without intermediate representation so as to carry out digital circuit design.

The invention provides a power load prediction method and device based on LoRA and GPT models. The method comprises the following steps: step 1, converting original power grid load time sequence data into a sample format which can be trained by a model; 2, configuring a model architecture, and taking a GPT model as a backbone network; 3, configuring a parameter fine tuning strategy, wherein the parameter fine tuning strategy comprises a partial unfreezing strategy and a low-rank adaptation strategy; 4, training and optimizing the model; 5, performing prediction, and inputting the normalized input sequence into the trained prediction model; and obtaining a load prediction result of a future specified time period output by the model. According to the method, a Transform time sequence basic model pre-trained on a large-scale corpus is adopted as a backbone network, a local unfreezing mechanism and a low-rank adaptive technology are introduced to construct a parameter efficient fine tuning strategy, the load prediction precision is improved, and the consumption capability of a power distribution network on distributed photovoltaic power generation is promoted.

This disclosure relates to machine learning models for performing atmospheric correction on an input image. To train a machine learning model to perform atmospheric correction on an input image comprising atmospheric distortion, a processor applies a first trained machine learning model to a training image to determine a first output image, the first trained machine learning model being configured to perform atmospheric correction on an input image comprising atmospheric distortion. The processor applies a second machine learning model to the training image to determine a second output image, wherein the second machine learning model has a smaller model architecture than the first trained machine learning model. The processor trains the second machine learning model by minimising a loss based on the first output image and the second output image to determine a second trained machine learning model.

A method includes creating a document retrieval and large language model architecture including at least one vector database including vectorized data corresponding to one or more documents from one or more document storage locations and a large language model. The method also includes receiving a query to generate a report associated with a current project using the large language model. The method also includes returning, in response to the query, a relevant context generated using the at least one vector database. The method also includes generating and outputting, using the large language model and based on the relevant context, one or more portions of the report.

The invention discloses a rolling bearing intelligent fault diagnosis method based on data quality leading. The method comprises the following steps: (1) collecting a vibration signal of a rolling bearing; (2) carrying out preprocessing and data enhancement on the collected signals, specifically, (2.1) optimizing a quantitative mapping relation between a sampling length and a fault period number based on a bearing fault characteristic frequency and a periodic impact theory, optimizing a signal length to be 2048 sampling points, and setting a sliding window overlapping rate to be 30%; (2.2) constructing a five-dimensional aggressive data enhancement strategy covering Gaussian noise injection, amplitude scaling, time translation, random flipping and impulse noise disturbance; (3) carrying out fault diagnosis on the rolling bearing; and (4) according to an experiment result, selecting an optimal data quality optimization strategy and a model architecture to carry out rolling bearing fault diagnosis, and outputting a fault diagnosis result. By optimizing the data quality, the accuracy and generalization ability of rolling bearing fault diagnosis are improved, and the problem caused by insufficient attention to the data quality in the prior art is solved.

The invention discloses a cooperative medical prediction system for a heterogeneous data center, and relates to the field of intelligent medical treatment, and the system comprises a distributed client set which is used for carrying out the localized training based on a local private medical image and dose data; the centralized coordination node is used for managing and coordinating a federation training process and executing aggregation of cross-client parameters; the decoupling model architecture comprises a globally shared feature encoder and feature adapters unique to a plurality of clients; the alternative training scheduling module is configured to periodically switch between a global aggregation mode and a localization adaptation mode, and the adaptive aggregation weighting module is used for dynamically calculating and distributing the weight of each client in global aggregation based on the statistical difference between data distribution and overall distribution of each client. According to the scheme, the overall generalization performance and prediction stability of the model on heterogeneous multi-center data can be improved, and the performance difference between centers caused by unbalanced data distribution is relieved.

The present application provides a data processing method and apparatus. The method comprises: acquiring a natural language processing (NLP) training data set; inputting the NLP training data set into a first large language model to train the first large language model, the trained first large language model comprising a first language modeling parameter; acquiring the first language modeling parameter; obtaining first text data on the basis of the NLP training data set; and obtaining a first prompt word set on the basis of the first text data, the first language modeling parameter, and a second large language model, wherein the first prompt word set and the first text data are used for training a third large language model, and the first large language model and the second large language model are based on the same model architecture. The method provided by the present application can obtain the first prompt word set of which the data volume is much smaller than that of the NLP training data set. In this way, using the first prompt word set to train the third large language model can improve the speed and accuracy of large language model training.

The invention discloses a water turbine local turbulence field prediction method, and relates to the technical field of water turbines. The method comprises the following steps: S1, data acquisition and preprocessing; s2, sparse measurement point feature extraction; s3, low-dimensional representation of the turbulent flow field; s4, carrying out double-model joint training; and S5, performing real-time prediction. According to the method, the global turbulent flow field prediction can be realized only by using the pressure / speed time sequence data of a small number of sparse measuring points in the water turbine runner, and an expensive full flow field experiment table does not need to be built or a dense sensor array does not need to be deployed. Compared with the traditional experimental measurement (the equipment cost is more than ten million yuan), the total investment of the sensor and the computing equipment is only ten thousand yuan, and the cost is reduced by more than 99%; meanwhile, the installation difficulty of the sensor in a high-flow-speed and high-pressure area is avoided, the existing monitoring point layout of a hydropower station can be directly adapted, the technology landing feasibility is remarkably improved, and the model architecture does not depend on a specific water turbine type and can be adapted to different types such as a mixed-flow type and an axial-flow type by adjusting grid parameters and training data.

The invention discloses an automobile production line parameter monitoring and early warning method and system based on dynamic threshold and multi-dimensional comparison. The method comprises the following steps: completing deployment of a sensor device for data acquisition and construction of a data set; constructing a model infrastructure, wherein an input layer synchronously receives a process card parameter vector, an equipment state vector and a process-equipment association feature vector; a time sequence segmentation learning mechanism is introduced into the LSTM layer, and data fragments are divided according to the equipment operation cycle; the model processing layer adopts a dynamic weight adjustment mechanism, and the Attention layer firstly endows features at different moments and in different equipment states with basic weights, and then carries out real-time calibration according to the abnormal contribution degree of each feature in historical early warning data; the output layer calculates and outputs the probability value of the actual parameter meeting the process requirement through a Softmax function; and taking a probability value output by the model as a core, establishing a hierarchical early warning response mechanism and matching a standardized process. And long-term accurate monitoring and efficient early warning of the parameters of the automobile production line are realized.

The invention discloses an artificial intelligence robot control system fused with a world model architecture, and the system comprises a control main system which comprises a shared multi-mode backbone network module, a strategy head, and a world model head. A visual encoder, an ontology perception encoder, a text encoder and multi-mode fusion are integrated in the shared multi-mode backbone network module, and the strategy head is used for generating a current action instruction. According to the method, a self-supervision signal provided by a world model is used as an additional training constraint, the dependence on large-scale teaching data is reduced, and a strategy head generates actions based on representation rich in physical dynamic information, so that the device has the advantages that the decision is more stable when facing environmental noise or uncertainty; error accumulation in a long-range task is remarkably reduced, and compared with a traditional open-loop strategy model, the method has the advantage that the generalization ability of training out-of-distribution scenes is remarkably improved.

The invention relates to the technical field of model training, in particular to an AI algorithm model training management system, which comprises the steps of analyzing a data value association matching degree in an AI algorithm model training process based on training data information and model deployment environment information of an AI algorithm model in the model training process; in combination with training configuration information and model architecture information of the AI algorithm model in the model training process, the effective utilization degree of computing resources in the AI algorithm model training process is analyzed; evaluating the training effect of the AI algorithm model based on the data value association matching degree and the effective utilization degree of the computing resources; according to the training effect evaluation result, the training scheme of the AI algorithm model is adjusted, the reliability and efficiency of the training process of the AI algorithm model can be remarkably improved, and the high performance and high robustness of the AI algorithm model in a real scene are ensured.