30 results about "Model building process" patented technology

The present application provides a CAD model analysis method and apparatus, a device and a storage medium, which relates to the technical field of industrial manufacturing. The CAD model based on a BRep representation is transformed into a graph structure. The graph structure is input into an analytical model to obtain a modeling sequence of the CAD model. By transforming the CAD model with a BRep representation into the graph structure, the analytical model learns the modeling feature in the graph structure, and the modeling feature type and the feature parameter are obtained, so as to enable the prediction of the model building process of the CAD model and facilitate to subsequently make parametric re-editing of the CAD model

The invention provides a method for automatically deploying artificial intelligence models, which simplifies a model building process through systematic data preprocessing, model selection, parameter optimization and performance monitoring mechanisms, and dynamically updates or switches models in an application environment to maintain overall prediction performance at the best state while improving the performance of the model in multiple application environments.

The invention discloses a new method for constructing an oil reservoir simulation agent model based on transfer learning, and relates to the technical field of petroleum engineering, and the method comprises the steps: building a fine-scale oil reservoir model; constructing a high-fidelity data set; carrying out scale upgrading and coarsening on the oil reservoir model; constructing a low-fidelity data set; constructing a deep learning model; training a pre-training model by using the low-fidelity data; finely tuning the pre-training model by using the high-fidelity data; and predicting output parameters. According to the method, it can be guaranteed that under the condition that a small amount of high-fidelity training data is used, oil reservoir multi-term flow under the variable well control condition can be accurately predicted, dependence on the high-fidelity training data in the agent model building process is effectively reduced, and the agent model building time is shortened on the premise that prediction precision is not affected. According to the method, the embedded module is introduced to process the time-varying well control parameters, the problem that matrix element parameters corresponding to well positions are difficult to effectively diffuse to the whole output result through finite times of convolution is solved, and the global influence of the well control parameters on the output result of the proxy model is improved.

The invention discloses a building information model automatic construction method and system and a storage medium. Multi-source data are collected and preprocessed to form input data in a unified format; respectively extracting two-dimensional and three-dimensional features by using a multi-branch architecture of a convolutional neural network, and comprehensively outputting feature information of the building component in a feature fusion layer; and calling a modeling software interface to convert an identification result into corresponding components, and sequentially assembling the corresponding components into a complete building information model. In order to ensure the accuracy of the model, the method introduces a repairing mechanism based on geometric constraint to adjust gaps and overlapping between components, meanwhile, the volume error rate and the position deviation of the model are verified through statistical analysis, if the volume error rate and the position deviation exceed threshold values, error data are fed back to a network to be retrained, and closed-loop optimization of the model construction process is achieved. And the automatic feature extraction capability of deep learning is combined with a BIM software modeling mechanism, so that the modeling efficiency and precision in a complex scene can be remarkably improved.

The invention provides a data insight method and system based on an AI large model, and the method comprises the steps: obtaining business data, and carrying out the preprocessing; constructing a data insight model based on a big language AI model, in the construction process, performing fine adjustment on the big language AI model by adopting an adaptive attention fusion loss function, and in the fine adjustment process, introducing a reinforcement learning mechanism to guide the model to generate an insight report meeting service requirements by taking a service target as a guide when the model analyzes data; and inputting the business data into the data insight model, carrying out deep analysis to obtain an analysis result, and generating a high-quality insight report according to a business optimization strategy obtained by reinforcement learning by taking a business target as a guide in the analysis process. According to the method, the deep analysis capability of the large language AI model, fine adjustment of the adaptive attention fusion loss function and guidance of a reinforcement learning mechanism are combined, so that the data analysis depth and accuracy of the model are improved, the service guidance quality is enhanced, and the quality of an insight report is improved.

The embodiment of the invention provides construction of a multi-working-condition tool wear monitoring model and a wear monitoring method, system, equipment and medium thereof, and belongs to the field of neural networks. The method comprises the steps that based on a Pytorch deep learning framework, an attention mechanism ECA is embedded into a two-dimensional convolutional neural network, and a subject network of a multi-working-condition tool wear monitoring model is constructed; performing training and fine tuning on the subject network through a multi-source domain transfer learning pre-training and fine tuning strategy; bayesian optimization is applied to the convolution kernel size in the pre-training stage and the hyper-parameter optimization process in the fine tuning stage, and a multi-working-condition tool wear monitoring model is obtained. The generalization ability of the tool wear monitoring model under variable working conditions can be improved by learning information related to the tool state in signals under different working conditions by using transfer learning. And a pre-training-fine tuning strategy is introduced into a model construction process, so that the adaptability of the model to a new environment and the construction efficiency of the model are improved.

The invention discloses an MCMC acceleration method and accelerator based on adaptive data sub-sampling, which are applied to Bayesian modeling, and the acceleration method comprises the following steps: generating a candidate parameter sample, and calculating the upper bound of the energy difference between the candidate parameter sample theta'and the current parameter sample theta i-1; generating the number B of small-batch quantum sampling according to the upper bound of the energy difference bound and the current hyper-parameter x; sampling according to the number B, and updating the acceptance probability likelihood LH of the small-batch data set and the theta'according to the acceptance probability of the data points; calculating an acceptance probability alpha by using the finally accumulated LH; accepting or rejecting theta'as a parameter sample of the current iteration of the model according to alpha; dynamically adjusting the hyper-parameter x every fixed iteration times; and repeating the process to obtain a final parameter sample theta Ns of the Bayesian model. According to the method, the optimal balance between the sampling speed and the sample efficiency of the parameters in the model building process can be realized.

The invention discloses a reasoning unloading method and system for a large model agent in an edge computing scene and a medium. The method comprises the steps that a to-be-processed target large model task is input into a pre-constructed Transform-PPO model to obtain a corresponding target unloading scheme used for decomposing the target large model task into a plurality of target sub-tasks and determining corresponding unloading decisions, and the model building process comprises the step of obtaining a reasoning chain of a directed acyclic graph from the large model task through a Transform network; generating an unloading scheme comprising an unloading strategy of each subtask of the inference chain based on a PPO algorithm, wherein the unloading strategy is an unloading decision of the subtask and a probability of the unloading decision; and punishing and / or rewarding the unloading strategy for increasing and / or reducing the total delay in the last round of training to obtain an unloading scheme in the current round of training. According to the method, a better reasoning unloading scheme of the large model task can be obtained, the processing efficiency of the large model task is improved, and the equipment energy consumption is reduced.

The present invention proposes a digital twin model construction method, which belongs to the field of artificial intelligence technology, including collecting historical data, establishing a historical database, extracting samples from the historical database through a dynamic extraction algorithm, and using the samples to build a digital twin model; performing data correction on the generated digital twin model in the model prediction stage; using the data-corrected digital twin model to predict the real-time data stream and obtain a target data set; recording parameter and data changes in the construction of the initial digital twin simulation model; and being able to reasonably allocate model optimization resources during the model construction process, and correspondingly optimize the constructed digital twin model, thereby improving the accuracy of the model. At the same time, in the model prediction stage, the prediction process of the model is tested and corrected, which can improve the accuracy of the model prediction.

The invention relates to the technical field of building environment and energy utilization, in particular to an air conditioning system cooling load prediction method and related equipment, and the method comprises the steps: firstly obtaining related data of an air conditioning system, inputting the data into a pre-trained improved LSTM load prediction model, and obtaining a cooling load prediction result; the construction process of the improved model comprises the steps that historical air conditioner system related data is obtained, and a data set is formed through screening by means of a random forest algorithm; constructing an LSTM load prediction model, and updating hyper-parameters in the model based on a dung beetle optimization algorithm; performing iterative training on the LSTM load prediction model by using historical cold load data in the data set to obtain an optimal hyper-parameter; and finally, the obtained optimal hyper-parameter is updated to the LSTM load prediction model, so that construction of the improved LSTM load prediction model is completed for subsequent cold load prediction.

The invention relates to an ISA-PSO-RVM fault power failure prediction model based on a simulated annealing-particle swarm optimization algorithm, and the model building process comprises the steps: data collection: collecting the historical data of fault power failure, including fault time, fault type, fault position and influence factors; model construction: determining an RVM model structure; selecting proper kernel functions and parameters, and constructing a basic framework of the RVM model; sA-PSO algorithm optimization: initializing a particle swarm, evaluating fitness, updating particle positions and speeds, carrying out iterative search until a preset number of iterations is reached or a stop condition is met, and outputting an optimal solution: selecting a particle with an optimal fitness value as the optimal solution, so as to obtain optimized RVM model parameters; and training the fault power failure data by using the optimized RVM model, and establishing a mapping relation between the fault power failure and the influence factors.

The invention relates to the technical field of dynamic design and predictive design, and discloses a super-model automatic modeling method, which specifically comprises the following steps of: 1, geometric model cleaning: performing geometric cleaning on an original structure model, eliminating geometric errors in a modeling process, and obtaining a geometric model; according to the automatic super-model modeling method, the number of iterations in the model building process can be remarkably reduced, manpower and computing resources are saved, the convergence and precision of the model are ensured through step-by-step refinement and correlation analysis, the reliability of the model is improved, modal matching is carried out through the modal confidence criterion (MAC), modal analysis of the model is more accurate, and the modeling efficiency is improved. According to the method, the dynamic characteristics of the structure can be accurately predicted, the convergence model can be rapidly determined by setting a clear convergence judgment standard, meaningless calculation is avoided, the working efficiency is improved, the finally determined super model can meet the double indexes of precision and efficiency, and a powerful tool is provided for dynamic design and prediction.

This invention discloses a method, system, and medium for inference offloading of large model agents in edge computing scenarios. The method includes inputting the target large model task to be processed into a pre-constructed Transformer-PPO model to obtain a corresponding target offloading scheme for decomposing the target large model task into multiple target sub-tasks and determining the corresponding offloading decisions. The model building process includes using a Transformer network to obtain an inference chain of a directed acyclic graph of the large model task; generating an offloading scheme based on the PPO algorithm, which includes offloading strategies for each sub-task, where the offloading strategy is the offloading decision and the probability of the offloading decision for each sub-task; and obtaining the offloading scheme for the current training round by penalizing and / or rewarding the offloading strategy that increases and / or decreases the total latency in the previous training round. This invention can obtain a better inference offloading scheme for large model tasks, improve the processing efficiency of large model tasks, and reduce device energy consumption.

The invention relates to a multi-axial fatigue prediction method and medium based on non-linear coupling of normal and shear strain energy, and the method comprises the steps: calculating the normal strain and tangential strain of each group of experimental results at each angle under a loaded condition by using the fatigue experimental results of a to-be-predicted material, finding out the plane where the maximum shear strain amplitude of the maximum normal strain is located, and calculating the maximum shear strain amplitude of the maximum normal strain; taking as a critical surface; calculating each stress-strain component on the critical surface; acquiring material and shear performance parameters of a to-be-predicted material, and predicting by adopting the fatigue prediction model in combination with each stress-strain component; the model building process comprises the following steps: performing nonlinear coupling on normal strain energy and shear strain energy, obtaining a fatigue damage parameter by combining an average normal stress correction coefficient and an average tangential stress sensitivity coefficient, and obtaining a fatigue prediction model by combining the fatigue damage parameter with a fatigue damage equation in a shear form. Compared with the prior art, the method has the advantages of being high in fatigue life prediction precision, suitable for proportional and non-proportional loading, high in safety and the like.

This invention provides an intelligent modeling method and system for steel performance prediction based on LLM (Liquidity Management Model). The system includes: receiving and parsing user instructions through a natural language understanding module to generate a standardized task description; automatically generating or matching an automated model building process based on the task description through a planning and execution module; providing decision support based on a knowledge base through a knowledge recommendation module during task planning or execution; and automatically scheduling functional modules such as data access, feature engineering, and model building to complete the modeling task and output the results, based on the recommended information, through the planning and execution module. The system accordingly includes various functional modules implementing the above method. This application achieves end-to-end automated modeling through natural language interaction, automated task planning, and intelligent knowledge recommendation, significantly improving modeling efficiency, lowering the technical threshold, and enhancing model quality and versatility.

The invention provides a maintenance tool for a future state power grid model and a construction method thereof. The maintenance tool comprises a ground state model storage module, a time sequence topology change operator library, a model construction engine and a simulation file generation module, wherein the ground state model storage module is used for storing a ground state detailed model of a power grid. According to the method, a linkage mechanism of analysis context and sequential topology change operators is established, firstly, a research area is determined according to the analysis context, operators outside the research area are pre-screened based on electrical distance and sensitivity analysis, and invalid changes with the influence on the research area lower than a threshold value are eliminated. According to the method, unnecessary detailed modeling of the whole network topology irrelevant to the analysis target is avoided, the target performance of the model construction process is remarkably enhanced, it is ensured that the computing resources are accurately used for constructing the model part strongly relevant to the analysis intention, and the pertinence and the computing efficiency of model construction are improved.

The application discloses a descriptor screening and crystal material physical property prediction method based on material genetic engineering, and the method comprises the following steps: collecting crystal descriptor by calculating and collecting part of crystal material target property data, then carrying out cleaning and comprehensive screening on the descriptor by virtue of variance information, four correlation coefficients and a random sequential feature selection method, then automatically optimizing and adjusting the hyperparameter setting in the model building process by using intelligent optimization thinking, finally applying the trained optimal model to the physical property prediction of the crystal material to obtain the prediction result, and overcoming the non-standard in the model building process of the material physical property prediction model, and improving the learning efficiency and prediction effect of the model.

The invention discloses a method and device for establishing a specific field modeling environment. The method comprises the following steps: constructing a three-layer abstract system for specific field modeling based on a meta-meta model; generating a domain meta-model containing entity class definition, relation class definition and enumeration class definition through an automatic instantiation rule, and configuring attributes, representation forms and constraint relations; packaging the meta-model and the constraint thereof into standardized framework data with a unique identifier; loading the frame data by a runtime analysis module, dynamically generating a visual domain model editor, and enabling a user to complete model construction conforming to constraints; triggering increment verification based on an operation-level event in a model construction process; when the meta-model is updated, executing difference comparison to identify affected model elements and giving repair suggestions; and the runtime expansion of the modeling capability is realized through a plug-in mechanism.

The invention discloses a small target detection model and method based on anchor frame focusing loss distribution correction, and relates to the technical field of target detection. According to the method, an overall framework of an AFL-DETR model for end-to-end small target detection is provided, a novel loss function specially customized for small targets is provided in the model building process, the area can be deduced by extracting position information from an object query, the area is incorporated into the loss function, contribution of the size of each target is adjusted, and the detection accuracy of the AFL-DETR model is improved. Therefore, the accuracy of small target detection is improved. In addition, an information enhancement module is also introduced, and the module fuses the preprocessed image details and the shallow feature information. And then, the features are further extracted and fused to obtain cross-fusion information, so that context semantic information suitable for the small target is generated.

The invention relates to the technical field of building three-dimensional data processing, in particular to a BIM-based building model data accurate conversion processing method. According to the method, the local density feature and the local curvature of each voxel point in the three-dimensional point cloud are analyzed, the influence degree of construction site noise on each voxel point is obtained based on the two features, and the filling necessity of each empty voxel block is obtained based on the overall influence degree distribution of neighborhood non-empty voxel blocks in a second neighborhood range of the empty voxel blocks. Whether the empty voxel blocks need to be filled or not is judged according to the filling necessity, and then a final filling result is obtained for BIM model construction and data conversion. According to the method, the finally constructed BIM building model is more accurate, the accuracy and adaptability of the model construction process under the interference of construction site environmental factors are improved, and data conversion is more accurate.

This application discloses a discrete element rapid modeling method suitable for large-capacity regularly arranged structures, which realizes the application of discrete element technology to large-capacity geotechnical structure modeling, especially multi-layer system structure model. In the model construction process, the large-capacity model is first divided into several sections by splitting and then merging, which improves the efficiency of skeleton structure construction; then, through the discrete element search area control algorithm, the traversal range of particle units is narrowed in a targeted manner, which greatly improves the efficiency of mixture model construction while ensuring the accuracy of model calculation. This method expands the application scope of discrete elements and fills the application gap of this technology in large-capacity models.

The invention aims to overcome the defects that a geometric model cannot integrate functions, is indefinite in semantics, is poor in universality and the like in a digital twinning technology and intelligent robot establishment process, and discloses a multi-joint robot digital twinning control method based on a body Petri net, which integrates five parts: simulation task establishment, virtual robot import and Petri net model modeling. Data communication and state updating, and model evaluation; specifically, the motion simulation control part is used for simulating the operation process and matters needing attention of equipment; in a digital twinning control stage, establishing an ontology coloring Petri net model; in the entity robot communication link, Kalman filtering is combined with a Petri network, and the filtering state is updated in real time; according to the method, the model building process can be simplified, traditional digital twinning function modules are inherited, the universality and expansibility of the built model are achieved in combination with the ontology Petri net, and the application value of the control model is improved.

The invention provides a decision model correction method and device, equipment, a storage medium and a program product, relates to the technical field of computers, and is used for reducing the subjectivity of decision model correction, reducing the complexity of a model construction process and ensuring the consistency of decision model correction. The method comprises the following steps: under the condition that a new feature is introduced into a decision scene, performing correlation analysis on the new feature and at least one decision index of a decision model to obtain at least one correlation coefficient; constructing a target weight matrix based on preset adjustment intensity, the at least one correlation coefficient and an original weight matrix of the decision model; and correcting the decision model based on the target weight matrix.

This invention relates to the field of computer-aided design technology, specifically to modeling methods, devices, and electronic devices. The method includes acquiring features and constraints from a target model document and obtaining the dependencies between features, where the features are geometric bodies in the target model involved in constraint solving; establishing a directed graph of the dependencies between features; grouping the constraints based on the directed graph to determine a sequence of constrained features; and performing constraint solving sequentially according to the order of the constrained feature sequence to determine the solution results, thereby establishing the target model. By establishing a directed graph of dependencies between features and grouping constraints, a hybrid solution of two-dimensional and three-dimensional constraints is achieved. This supports both two-dimensional sketch constraints and three-dimensional rigid body constraints during the target model building process, thus avoiding switching between the two environments during modeling and improving modeling efficiency.

The invention discloses a modeling and predicting method based on machine learning azeotrope system judgment, which is used for predicting whether a binary mixture forms an azeotrope system or not. The method comprises the steps of data set feature engineering construction and azeotrope system prediction model construction, the data set feature engineering construction comprises the steps of data cleaning, mixture input feature construction, feature conversion and the like, and the prediction model construction comprises the steps of data set division, model selection, model hyper-parameter optimization and the like. The model hyper-parameters needing to be optimized comprise the number of hidden layer nodes, the Dropout probability, the number of iterations, the batch size and the like. In the model construction process, through data preprocessing, feature construction, model parameter optimization and model evaluation, the prediction accuracy and the model generalization ability are improved. The method solves the problems of long time consumption, high cost, limited prediction range and the like of a traditional experimental analysis method, and has a wide application prospect.

The invention discloses a GNSS receiver observed quantity noise estimation method. The method comprises the following steps: S1, acquiring a reference coordinate value and observed quantity of a receiver; s2, initializing parameters; s3, obtaining a current observed quantity noise estimation value; s4, obtaining an observed quantity correction value; s5, obtaining a position estimation value of the receiver; s6, obtaining a position coordinate error of the receiver; s7, judging a skipping step according to the current position coordinate error of the receiver; s8, judging whether an inferior solution accepting condition is accepted or not; s9, assigning an optimal estimated value of observed quantity noise; s10, judging whether the number of iterations exceeds the limit or not; and S11, judging whether to terminate the search of the optimal estimated value of the observed quantity noise. The method does not depend on observed quantity noise modeling, randomly searches and screens observed quantity noise in a multi-dimensional space, avoids a noise modeling process, obtains a global observed quantity noise estimation value, and estimates part of observed quantity noise which is not accurately modeled under the condition that part of error sources are accurately modeled.

The invention provides a numerical calculation method and a numerical calculation device for a high-speed flywheel in a real-time digital simulator. The numerical calculation method comprises the following steps: establishing a numerical calculation verification platform for a single-machine energy storage system of the high-speed flywheel in the real-time digital simulator; then performing numerical modeling on the high-speed flywheel single-machine energy storage system on the high-speed flywheel single-machine energy storage system numerical calculation verification platform to obtain a modeling equation of the high-speed flywheel single-machine energy storage system, and performing equivalent modeling numerical calculation on the modeling equation of the high-speed flywheel single-machine energy storage system by adopting a trapezoidal integral algorithm and a Taylor series algorithm; obtaining a modeling equation calculation result of the high-speed flywheel single-machine energy storage system, wherein the calculation result reflects the performance characteristics of the high-speed flywheel single-machine energy storage system; the problems that numerical calculation is unstable and the model building process is complex in the simulation model building process are effectively solved, and the research and development period of related control and protection equipment of the flywheel energy storage system is shortened.

The invention discloses a stool dragon mathematical model calculation method, which comprises the following steps of S1, establishing a stool dragon motion mathematical model; s2, on the basis of the model in S1, setting an algorithm under the condition that stools are stopped due to mutual collision and cannot continue to enter, and solving; the method has the advantages that the variable step size search algorithm based on dichotomy is adopted, the optimal solution can be efficiently approached in the global range, fine optimization can be achieved in the local area, and in the model building process, the optimal solution can be accurately obtained, so that the optimal solution can be accurately obtained. Double mapping is constructed by using an idea of converting a curve into a straight to obtain a correlation between variables, an iterative solution idea is provided, and a vector method is used for participating in modeling, so that the calculation amount is reduced, and the programming difficulty is reduced.

The invention relates to an artificial intelligence model construction method and device based on deep learning, and the method comprises the steps: obtaining a data set, carrying out the coordinate marking of the data set, and obtaining a training sample; loading a feature architecture of the pre-training model, and adding an expansion architecture on the basis of the feature architecture; freezing the feature architecture, and adjusting the weight of the extended architecture to a set value; and training the pre-training model by using the training sample according to the learning rate to obtain a target model. By adopting the method, high calculation cost and time for training the model from zero can be avoided, the model accuracy and efficiency of a specific task are improved, the method is also suitable for a model training condition lacking a large amount of data, the model can be further optimized on the basis of not damaging original characteristics, and the method has high practicability. And the risks of over-fitting and model performance reduction in the model building process are avoided.