634 results about "Reduced model" patented technology

The invention discloses a strategy question-answering system and method based on a vertical domain large model, and the method comprises the steps: S1, receiving the natural language input of a user, and retrieving an open source data set to obtain an original data set with the comprehensive similarity meeting the requirements; s2, performing necessary data cleaning and preprocessing on the original data set to obtain a training set; s3, performing fine tuning on the pre-processed basic large language model by adopting the training data set, and introducing a low-rank structure to modify a weight matrix to obtain a final model; s4, the problem enters a final model for post-processing to generate prediction output; the system takes a large language basic model subjected to fine tuning training as a core, introduces a low-rank structure to perform fine tuning on model parameters, reduces parameter quantity required by training while keeping model performance, reduces model complexity, reduces an overfitting risk, improves generalization ability, supports deep fusion of a specific field knowledge base and universal field data, and has a wide application prospect. And the accuracy and correlation of questions and answers are improved.

The embodiment of the invention provides a task processing method, a model training method and an intelligent dialogue and model training platform. The task processing method comprises the steps of obtaining initial prompt information corresponding to a target task; the execution process of the target task is disassembled based on the initial prompt information through a natural language model, multiple target sub-tasks are obtained, target prompt information corresponding to the target task is generated, and the target prompt information comprises prompt information corresponding to the multiple target sub-tasks; and determining target task models of specifications corresponding to the plurality of target sub-tasks, and correspondingly executing the plurality of target sub-tasks by using the plurality of target task models based on the prompt information corresponding to the plurality of target sub-tasks to obtain task execution results of the target tasks. By disassembling the execution process of the target task and calling the target task models of different specifications, computing resources can be optimized, the model calling cost can be reduced, and the system efficiency can be improved.

The invention discloses a design method of a coreless stator axial magnetic flux permanent magnet motor. The method specifically comprises the following steps: firstly, determining a basic structure, a permanent magnet structure and performance demand parameters of the coreless stator axial magnetic flux permanent magnet motor; determining stator and rotor structure parameters of the motor according to the performance requirement parameters of the motor; establishing a simplified model of an internal magnetic circuit of the motor, determining excitation magnetic potential of the motor, and establishing a finite element dynamic simulation model of the coreless stator axial flux permanent magnet motor based on a Maxwell equation; and finally, comparing the finite element dynamic simulation result with the motor performance demand parameter, if the finite element dynamic simulation result does not meet the requirement, correcting the motor design structure parameter and the electromagnetic parameter, and re-determining the stator and rotor structure parameter of the motor until the finite element dynamic simulation result of the motor meets the motor performance demand parameter. The design cost of the coreless stator axial magnetic flux permanent magnet motor is reduced, the design period is shortened, and the design efficiency and the design precision are improved.

The invention relates to an aquatic environment simulation monitoring method and device for evaluating the influence of pollutants, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring environmental parameters and dividing the range of a fine model based on the environmental parameters; constructing a simplified model based on the environmental parameters in the non-fine model range; constructing a fine model based on the environmental parameters in the fine model range; and outputting and displaying a corresponding simulation result based on the simplified model and the fine model. By the adoption of the method, through multi-scale coupling of the simplified model and the fine model, it can be guaranteed that the diffusion trend of pollutants can be rapidly simulated in a large range, detailed analysis can be conducted in the local key area through the high-precision model, and therefore the overall simulation efficiency and precision are improved.

The embodiment of the invention provides a method for reducing model output illusion based on knowledge retrieval. The method comprises the following steps: constructing a knowledge base for storing knowledge entries; receiving a question text input by a user, converting the question text into vector representation, performing matching retrieval on the vector representation in the knowledge base, and retrieving and outputting candidate knowledge entries; obtaining an output answer text of the question text generated by the large model, comparing the candidate knowledge entries with the output answer text, and judging whether the candidate knowledge entries violate the output answer text or not; and performing illusion detection on the output answer text, correcting the output answer text according to an illusion detection result, and outputting an answer corresponding to the question input by the user. According to the method, real-time or near-real-time verification and intervention can be carried out on model output, and illusion is inhibited from the source or the early stage.

The positioning error of a feeding system obviously influences the machining precision, especially under the high-speed and high-load working condition. An existing positioning error compensation system often faces the problems of high calculation cost and complex sensor configuration and is difficult to be widely applied to the industry. Therefore, a simplified model needs to be established and an economical and efficient sensor integration and real-time compensation scheme needs to be developed urgently. Therefore, the invention discloses a dynamic positioning error compensation system of a numerical control machine tool feed driving system, and provides an error suppression system fusing a mechanism driving positioning error prediction model and a real-time compensation strategy. The positioning error prediction model is constructed as a hyperbolic cosine function of temperature and nut position, and the residual error-based dynamic compensation model can effectively process unmodeled errors and system dynamic changes. The dynamic positioning error compensation system shows excellent adaptability and precision under the dynamic working condition, and an extensible practical solution is provided for positioning error suppression in the high-precision manufacturing fields of aerospace, automobiles, electronics and the like.

The invention relates to the technical field of edge computing, distributed artificial intelligence and multi-agent collaboration, in particular to a cross-device multi-agent edge collaborative reasoning system and method, and the system comprises a cloud coordination module which is responsible for maintaining main model parameters, grading based on importance scores, and receiving edge device data to optimize a model; the edge management module communicates with the cloud, receives parameter grading data, allocates parameter subsets according to equipment resources, integrates inference experience and returns the inference experience to the cloud; the equipment execution module receives the parameter subset, executes lightweight reasoning, records experience and sends a cooperation request; the distributed knowledge distillation module realizes multi-level heterogeneous distillation, differential parameter selection and adaptive parameter compression; the collaborative reasoning module constructs a dynamic collaborative topology, allocates subtasks and integrates multi-device results; according to the system, the edge resource demand is obviously reduced, the model volume is reduced by 75-95%, 91% reasoning precision is kept, and efficient collaborative reasoning is realized.

The invention discloses a transmission tower vulnerability analysis method and system based on seismic simulation, and belongs to the technical field of electric power facility disaster prevention, and the method comprises the steps: correcting bedrock seismic oscillation parameters based on equivalent shear wave velocity, and generating a high-precision earth surface PGA distribution diagram; automatically positioning a key node layer through eigenvalue decomposition, reducing the order of the three-dimensional model of the power transmission tower into a two-dimensional series multi-degree-of-freedom model, and outputting a layer mass vector, a condensation stiffness matrix and rod piece bending stiffness; performing time-history analysis by using the corrected PGA-driven reduced-order model to generate an earthquake damage probability; the multi-source remote sensing data and the PGA are fused, the landslide space probability is output through a deep learning model, and the impact strength and the landslide kinetic energy are calculated to judge the landslide damage probability; and based on the dual-threshold condition triggering probability union set, calculating a comprehensive damage probability, and generating a three-dimensional vulnerable curved surface. According to the method, the problems of high missing report rate and low calculation efficiency of single disaster assessment are solved, and minute-level accurate early warning of the composite disaster risk of the power transmission tower is realized.

The invention discloses a sudden drought identification method and system based on a space-time double-branch fusion model. The method comprises the steps that meteorological data are acquired and preprocessed; calculating a composite sudden drought index according to the obtained data; performing data dimension reduction on the obtained data through singular value decomposition (SVD); a deep learning model is adopted to construct a time branch model, a graph attention network GAT is adopted to construct a space branch model, and the time branch model and the space branch model are dynamically fused through a cross attention mechanism to construct a space-time double-branch fusion model; according to the method, data dimensions are compressed and model complexity is reduced by fusing multi-source variables and combining singular value decomposition (SVD), meanwhile, geographic neighborhood weights are dynamically learned by adopting Transforme and based on a graph attention network (GAT), dynamic fusion of spatial-temporal characteristics is finally realized through a cross attention mechanism, a sudden drought recognition result is generated, and the method has the advantages of being high in robustness, high in accuracy and high in reliability. The limitation of a traditional method on nonlinear feature capture, space-time modeling splitting and generalization ability is broken through.

The invention is suitable for the technical field of computer vision and deep learning, and provides a multi-scale feature fusion concrete defect detection method based on improved SAM, and the method comprises the steps: obtaining a concrete defect image data set; preprocessing the image data set and dividing the image data set into a training set, a verification set and a test set; the YOLOv9 is trained to automatically detect concrete defects; constructing an improved SAM model, replacing a prompt encoder of the SAM with YOLOv9 and edge detection, and modifying a mask encoder of the SAM; the training effect of the model is evaluated through the four indexes of the accuracy rate, the recall rate, the F1 value and the intersection-to-union ratio, and the improved SAM model is optimized according to the training effect. By improving the structure of the SAM, the detection capability of the model on small defects can be enhanced, and the detection accuracy and efficiency are improved. In addition, according to the method, the complexity of the model can be reduced, and the training efficiency and generalization ability of the model are improved.

The invention discloses a high-precision electrical calibration method and system for a wedge flowmeter. The method comprises the following steps: step 1, data acquisition and preprocessing; step 2, dynamic working condition identification; 3, updating the self-adaptive model; 4, multi-source data fusion calibration is carried out; and 5, performing closed-loop verification and compensation. According to the invention, multiple types of signals are collected in real time through the multi-dimensional sensor array and de-noised to construct a data set, so that comprehensive and accurate data collection is realized, and a good basis is provided for calibration; carrying out dynamic working condition identification by utilizing a support vector machine, triggering model updating, and endowing calibration with dynamic self-adaptive capability; model parameters are updated by adopting a recursive least square method, compensation items are established, and the characterization capability of the model is enhanced; a calibration coefficient is generated by means of a data fusion algorithm, and the calibration precision is improved; through verification and compensation of a digital twinborn model, traditional defects are effectively overcome, model transplantation errors are reduced, the measurement precision of the wedge flowmeter is improved under extreme working conditions, and production stability and product quality are guaranteed.

The invention discloses a machine learning model compression method and system based on federated learning and mean iteration, and belongs to the field of deep learning optimization, and the compression method comprises the following specific steps: I, a server initializes a group of basic models, and distributes the models to each client node; iI, updating model parameters by each client node by using local data, and aggregating the client parameters into a global model by the server side; according to the method, the network bandwidth burden and the risk of data privacy leakage can be reduced, effective fusion of training results of different clients can be promoted, the training quality of a global model can be improved, the overall performance and expandability of a system can be remarkably improved, meanwhile, model redundancy is reduced, and the calculation efficiency is improved; the method can further reduce the storage and calculation overhead, improve the compression efficiency, effectively maintain or even improve the precision of the compressed model, ensure the effectiveness of the model at different clients, reduce the communication overhead and calculation burden, and improve the overall efficiency of the system.

The invention requests to protect an unmanned aerial vehicle light-weight small target detection method based on a YOLO-GLL neural network, and the method comprises the following main steps: S1, constructing a YOLO-GLL target detection model based on YOLOv8s, and inputting a training set image to complete model training; s2, extracting different scale feature maps of the image through a backbone network containing a partially grouped multi-scale convolution module; s3, performing cross-level fusion on the feature map by using the lightweight multi-scale fusion feature pyramid neck network; s4, executing target classification and bounding box regression by adopting a lightweight shared detail enhancement detection head to obtain a lightweight high-precision model; and S5, inputting a to-be-detected unmanned aerial vehicle image, and realizing small target detection and identification in combination with the Soft-NMS and Shape-IoU post-processing strategies. The method has the advantages that (1) the detection precision is improved; (2) the parameter quantity of the model is reduced, and the deployment of resource-constrained equipment is adapted; and (3) a post-processing strategy is improved, the problem of missing detection of small targets in a dense shielding scene is relieved, and the robustness of a complex scene is enhanced.

The invention discloses a retrieval method and device based on a hybrid vector model and a readable storage medium, and relates to the field of artificial intelligence, and the method comprises the steps: firstly responding to a natural language query, and processing to generate a query vector; recalling candidate results from the index library based on the query vector, and screening to obtain a refined related information set; based on the refined information set, generating and executing an SQL, obtaining result data and generating a natural language answer; and outputting answers and result basis descriptions. According to the method, the retrieval accuracy is improved through mixed vector recall, the model reasoning complexity is reduced through multi-stage processing, the robustness is guaranteed in combination with a reflection mechanism, interpretable output meets the compliance requirement, the problems that retrieval is inaccurate, generation is unreliable and a result cannot be interpreted are effectively solved, and efficient automation of complex Text2SQL query is achieved.

The invention relates to the technical field of concrete structure damage detection, in particular to a post-fire RC structure burst damage three-dimensional segmentation method, which comprises the following steps: firstly, determining the type and three-dimensional characteristics of post-fire RC structure member surface concrete burst damage, and constructing a corresponding three-dimensional point cloud data set for network training and verification; and then based on a KP-FCNN network structure, a KPConv layer is improved and optimized, so that the detection and segmentation precision is improved, the model size is reduced, the reasoning time is remarkably shortened, the optimal segmentation precision of 82.3% is achieved under the working conditions of different damage degrees, automatic damage segmentation of the burst damage of the concrete structure after the fire disaster is realized, and the automatic damage segmentation of the burst damage of the concrete structure after the fire disaster is realized. And technical support is provided for subsequent damage three-dimensional quantification and deployment to an unmanned aerial vehicle system.

The invention relates to the technical field of die design, in particular to a method for acquiring a die surface deformation compensation value of a stamping die. The method comprises the specific steps that S1, a press table top deformation measuring system is adopted, and the deformation amount of a press under a certain load is collected; s2, establishing a simulation model of a measurement system by adopting Hypermesh software, and obtaining a simplified substitution model of a press table board and a sliding block by combining the collected experimental data through size optimization; s3, cleaning and grid division are conducted on the geometric structure of the mold, and accessory structures which do not affect the strength are removed; s4, combining the simplified model with the die grid model, adding load and constraint of a press machine, and performing stamping simulation; and S5, generating a die surface deformation contour map according to a simulation result, and obtaining a die surface deformation compensation value. Compared with the prior art, by combining a high-precision measurement technology and numerical simulation analysis, technical support can be provided for design and manufacturing of the stamping die, and progress and development of the die industry are promoted.

The invention discloses a QEM algorithm optimization method based on sharp feature degree and area weighting, and belongs to the technical field of image processing, and the method comprises the steps: S1, obtaining a to-be-simplified model, carrying out the simplification through a QEM algorithm, replacing the original folding cost with the optimization folding cost during the simplification, and obtaining a first simplified model; and S2, optimizing each vertex in the first simplified model based on a Laplacian operator to obtain an optimal simplified model. According to the method, on the basis of a traditional QEM algorithm, a new constraint factor is introduced to improve folding cost and generate optimized folding cost, model folding can be considered from more aspects, especially on geometric detail feature processing, model vertex optimization is additionally introduced in model rendering except for operation simplification, and therefore the model folding efficiency is improved. Low-quality long and narrow triangles can be eliminated to the greatest extent, so that the surface of the geometric model is smoother after rendering, and the overall quality of the model is improved.

The invention discloses a non-parametric probabilistic load flow calculation method considering complex uncertainty for a power system. The method comprises the following steps: firstly, constructing source-load nonparametric probability distribution based on a multivariate Gaussian mixture model; then, constructing an alternating current power flow model under a polar coordinate system in the power system, and performing linear expansion on the alternating current power flow model under the polar coordinate system in the power system at a mean point of source-load non-parametric probability distribution by adopting a first-order Taylor series approximation method to obtain a system operation simplified model based on alternating current power flow linearization of the power system; and finally, according to the linear transformation property of the Gaussian distribution, obtaining the probability distribution of the power flow distribution of the power system under each Gaussian component, and in combination with the total probability law, carrying out weighted summation on the probability distribution under each Gaussian component according to the weight to obtain the non-parametric probabilistic power flow probability distribution of the power system. The method provided by the invention has stronger generalization ability and accuracy, and has important reference significance for operation uncertainty quantification of the power system under complex uncertainty.

The invention discloses a multi-target prediction method, system and device for a financial time sequence and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: firstly obtaining financial time sequence data and auxiliary information, and carrying out the vector conversion and fusion to form a unified input representation; extracting features through a shared encoder, and generating a global feature vector; and inputting the data to a multi-task prediction output head to realize parallel prediction of the prediction tasks. According to the scheme, cross-task feature sharing is realized by using the shared encoder, the modeling collaboration is improved, and resource waste and prediction conflicts are avoided; through data fusion, the depiction capability of complex financial dynamics is enhanced; the generalization and robustness of the model are improved by adopting a lightweight output head, the method is suitable for diversified financial prediction scenes, and the prediction efficiency and accuracy are improved while the model complexity is reduced.

The invention provides a simplified modeling method and device for a complex mixed fuel combustion reaction kinetic model and a computer program product. The simplified modeling method comprises the steps of model coupling, primary model simplification by adopting a directed relation graph method based on error transfer and a species sensitivity analysis method, and optimization of the primary simplified model by taking an optimal pre-exponential factor, a temperature index and activation energy of a three-parameter modified Arranius equation as optimization objects. And the optimization model is simplified again. According to the method, a method of directly simplifying a complex model is adopted, and the model does not need to be manually disassembled and constructed; the method does not need deep reaction path analysis, does not need programming, is suitable for combustion simulation engineering practitioners without the basis of combustion reaction kinetics research, and is used for rapidly making a fuel mechanism model and carrying out reasonable verification.

The invention discloses a bedrock discrete fracture network model generation method based on multi-source drilling data. The bedrock discrete fracture network model generation method comprises the following steps: step 1, preprocessing multi-source data; step 2, recognizing a drilling fracture image, and clustering trace line segments recognized in a drilling image area into continuous fracture trace lines; step 3, fracture parameter statistics; step 3.1, calculating and determining geometric parameters of the fracture; step 3.2, performing statistics on a rock mass fracture network distribution model and parameters; 4, generating semi-determined fracture networks in batches; 5, multi-source data iterative analysis is carried out, and an optimization model is selected. Modeling is carried out through multi-source data cross comparison without depending on a single data source, and the accuracy of the model is improved. Automatic analysis is achieved through a digital method, manual operation is reduced, dependence on engineering experience is reduced, and large-scale application in the engineering scale can be achieved; and meanwhile, the determined crack of the drilling area and the uncertain crack of the drilling adjacent area are considered, so that the model uncertainty of grouting engineering modeling is greatly reduced.

The invention relates to the technical field of tunnel monitoring, in particular to a TBM tunnel weak broken zone jamming risk prediction and active reinforcement decision-making method. According to the method, the TBM shield pressure prediction sub-model and the multiple TBM feature data are used for prediction, Dropout operation is introduced, and model overfitting is reduced; according to the method, the deviation degree is utilized for risk grading, the risk grades can be rapidly and effectively obtained, different supporting modes are selected according to the different risk grades, and the problem that a reinforcement treatment method and the surrounding rock condition cannot be effectively matched under the condition that the machine jamming risk exists is solved; the system is simple in structure and high in applicability, a proper active reinforcement decision is made, resource waste is reduced, workers are assisted in completing tunnel construction, TBM equipment is prevented from being damaged, and the occurrence rate of engineering accidents is reduced.

The application provides a simplified modeling method, device and computer program product of a complex mixed fuel combustion reaction kinetics model. The simplified modeling method comprises the following steps: model coupling, one-time simplification of the model by using an error transmission-based directed relationship diagram method and a species sensitivity analysis method, optimization of the one-time simplified model by taking the optimal pre-exponential factor, temperature index and activation energy of the three-parameter modified Arrhenius equation as optimization objects, and re-simplification of the optimized model. The application adopts a method for directly simplifying a complex model, does not need to artificially disassemble and construct the model, does not need to perform in-depth reaction path analysis, does not need to perform program writing, is suitable for combustion simulation engineering practitioners without a combustion reaction kinetics research foundation, is used to quickly produce a fuel mechanism model, and is reasonably verified.

The invention relates to a humanoid robot multi-particle model prediction control method and device, and the method comprises the steps: simplifying a target humanoid robot into a multi-particle simplified model based on at least one target experiment task and mechanical structure parameters; based on the at least one target experiment task, determining a control task and a control constraint of each particle in the multi-particle simplified model; calculating an optimization variable of the target humanoid robot based on the control task and the control constraint; and taking the optimization variable as a reference trajectory, and carrying out trajectory tracking by utilizing whole body control to solve a torque instruction of each joint of the target humanoid robot, so that the target humanoid robot executes a corresponding target experiment task under the condition of meeting a preset balance condition. Therefore, the technical problems that in the related technology, a simplified model used for model prediction cannot fully utilize the redundant freedom degree of the robot, and it is difficult to adjust the body balance under the condition that the real-time performance and efficiency are met are solved.

The invention discloses a welding simulation heat source parameter calibration method and system based on a hybrid agent model, and belongs to the technical field of welding numerical simulation and optimization, and the method comprises the following steps: S1, problem definition and experiment design; s2, analyzing a product structure and constructing a joint model; s3, parameter space sampling; s4, performing simulation calculation; s5, constructing a mixed agent model; s6, proxy model optimization based on a genetic algorithm; s7, parameter verification and model updating; and S8, optimizing and outputting optimal parameters. According to the method, most time-consuming full-order finite element calculation is replaced by the proxy model and the simplified joint model, the parameter calibration process is shortened from several days or weeks to several hours, and the optimization efficiency is improved; a simplified model and a complete model are used for optimization, firstly, the simplified model is used for rapidly obtaining an optimal parameter approximate value, and the complete model is used for obtaining accurate parameters.

The invention discloses a real-time detection system and method for coating surface defects based on deep learning, and aims to improve cross-product-line detection adaptability, reduce model training cost and enhance detection stability under complex illumination through multi-scale feature extraction, feature decoupling and small sample adaptation technologies. The system obtains a coating surface image through an image acquisition module, a multi-scale feature extraction sub-module extracts low-level, middle-level and high-level features, the features are decomposed into domain invariant defect features and domain specific features through a feature decoupling sub-module, and interference features are suppressed through adversarial training. The small sample adaptation sub-module can realize rapid parameter adjustment and data enhancement by means of a small number of labeled samples, and the feedback control module triggers a production line repair instruction in real time and generates a quality report, so that the problem of false detection caused by cross-domain detection performance degradation, poor small sample adaptability and illumination interference is solved, the false detection rate of cross-product line detection is reduced, and the detection efficiency is improved. The sample demand is reduced, and the detection accuracy is improved.

A combination of two or more trained machine learning models can exhibit a combined accuracy greater than the accuracy of any one of the constituent models. However, this increase accuracy comes at additional computational cost. Cascades of machine learning models are provided herein that result in increased model accuracy and / or reduced model compute cost. These benefits are obtained by conditionally executing one or more of the models of the cascade based on the estimated correctness of already-executed models. The estimated correctness can be obtained as an additional output of the already-executed model(s) or could be determined as an entropy, maximum class probability, maximum class logit, or other function of the output(s) of the already-executed model(s). The expected computational cost of executing the model cascade is reduced by only executing the downstream model(s) when the upstream model(s) has resulted in an output whose accuracy is suspect.

The invention discloses a multi-variable measurement sensor state lightweight evaluation method based on knowledge distillation, and belongs to the technical field of electric digital data processing and multi-sensor data fusion. The method comprises the following steps: firstly, carrying out time synchronization and physical consistency constraint modeling on original data of multiple sensors, and extracting feature representation; a high-precision teacher model is trained at the cloud end, and the high-dimensional mapping relation of the sensor state is learned; intermediate features, soft output and uncertainty information of a teacher model are extracted to serve as distillation knowledge, lightweight student model training is guided, and effective migration of discrimination knowledge is achieved in combination with soft label constraint, feature alignment and an uncertainty guiding mechanism; and finally, compressing, quantifying and optimizing the student model, and deploying the student model to a vehicle-mounted end to realize real-time evaluation and dynamic updating of the states of multiple sensors. According to the method, the model complexity is greatly reduced while the evaluation precision is ensured through a knowledge distillation framework, and efficient and reliable state perception and fault-tolerant control support is provided for an intelligent driving system.

The invention discloses a flywheel system bearing assembly multi-dimensional rule reduction health state testing method, and relates to the technical field of flywheel system bearing assembly health state testing. Acquiring an actual measurement data set of the flywheel system bearing assembly; determining a confidence coefficient corresponding to the reference value set and the result, and constructing an initial health state test model based on a confidence rule base; constructing a multi-dimensional rule evaluation framework; performing composite activation on the reduced model; optimizing parameters in the model by adopting an optimization algorithm; and reasoning the model by using an evidence reasoning analysis algorithm to obtain a health test result of the flywheel system bearing assembly, and realizing an accurate health state test through the evidence reasoning algorithm.