6 results about "Sample space" patented technology

The application discloses an Optuna-XGBoost tunnel stability prediction-based hyperparameter optimization method and relates to the technical field of tunnel construction, and solves the problem that the existing technology is difficult to capture the complex deformation mode by using the posterior probability criterion as the deformation anomaly point strategy, is prone to misjudgment, and influences the accuracy and generalization ability of prediction; the application comprises the following steps: according to the proportion of real data and simulation data in the stability comprehensive index category and the joint fissure rock mass tunnel stability sample database, a training set and a verification set are divided to obtain a sample space; an XGBoost classifier is trained by using an Optuna Bayesian optimization framework based on TPE and the training set; category weights and data source weights are constructed and are distributed to the training samples; after multiple rounds of training, the performance effect is verified by using the verification set to determine the optimal hyperparameters of the XGBoost classifier; and the application improves the applicability and robustness of the prediction model under the condition of complex joint fissure surrounding rock.

The application discloses a kind of full-automatic process simulation method for product design, multiple-factor influence, multiple-target optimization, comprising:1, the data interaction between Creo Parametric and Ansys is established;2, parameterized model is established in Creo Parametric;3, parameterized model is imported into Ansys, meshing, case setting and result processing are carried out, and visual result is obtained;4, based on visual result, DOE experimental design is carried out in Desigin-Expert, and initial sample space is generated;5, initial sample space is imported into the parameter set of Ansys, parameterized simulation is carried out, and initial sample space target value is obtained;6, initial sample space target value is imported into Design-Expert, and target function between multiple factors and multiple targets is fitted;7, based on multiple-target optimization algorithm and weight setting, multiple-target optimization is carried out, and multiple-target optimization result is obtained;8, multiple-target optimization result is returned to Ansys and simulation verification is carried out.The application realizes the automation and process simulation under different structure parameters, and saves simulation time.

The invention relates to sampling training data for training a graph neural network for electronic design automation. A graph database stores (1) a plurality of graphs each representing a finished electronic design, with each graph containing components nodes with component features, and edges, wherein each component node represents a physical component. A graph clustering module assigns (3) a cluster from a set of clusters to each component node, wherein each cluster represents component nodes with common characteristics in topology and / or common component features. A cluster-based link sampling module samples (4) an edge from a sample space, wherein the sample space consists of all edges connecting clusters of component nodes in a selected graph currently selected among the graphs, and wherein a component node connected to the sampled edge is taken as a source component node, and wherein a subgraph is taken from the selected graph based on a receptive field around the source component node. Finally, a cluster-based receptive field sampling module picks (5) a cluster of component nodes that is connected to the source component node within the subgraph, reduces (6) the subgraph by dropping the picked cluster of component nodes, creates a corresponding reduced subgraph, and adds (7) the reduced subgraph along with the source component node and a target to a training set, wherein the target is a label of the picked cluster. These operations can be repeated iteratively (8), until every cluster has been picked, and / or recursively (9) within the reduced subgraphs, until only one cluster remains. Further embodiments provide a novel end-to-end training procedure for GNN-based recommender systems based on the efficient link sampling, which results in faster convergence during training, thus saving both compute resource costs and expert's time while waiting for new models. As a consequence, this enables the possibility to explore a wider range of models (e.g., a larger hyperparameter range) within given constraints.

The application discloses a ship type optimization method based on a partitioned hybrid proxy model, and aims at parameter optimization of an engineering problem; two-stage partition of an engineering parameter sample space is completed through I-DBSCAN clustering and Voronoi geometric division; sample balance is realized in combination with a Monte Carlo method and an improved SMOTE algorithm; adaptive sampling optimization is performed on a global Kriging model based on a G-EBM, local RBF models are constructed for each subspace, model weights are determined through error reciprocal weighting and boundary distance smoothing, the prediction results of the local RBF and global Kriging models are fused, and high-precision parameter analysis and optimization of the engineering problem are realized. The method solves the problems of unreasonable partition of the engineering parameter sample space, uneven sample distribution and large boundary prediction error, greatly improves the efficiency and precision of parameter optimization of the engineering problem, and is especially suitable for complex ship engineering parameter design scenes such as ship type optimization.

This invention relates to the field of composite sample processing technology in image defect detection, and provides a machine learning-based method for calculating the spatial efficiency of composite samples. The method includes: acquiring the current detection model, multi-source sample data, and corresponding contextual constraint data to construct composite sample spatial units; calculating the boundary gain, drift demand, coverage gap, confidence metric, and redundancy suppression of each spatial unit to obtain the spatial unit efficiency; performing action calculations under constraints of budget, real anchor points, and synthetic injection ratio; performing defect synthesis on the supplementary sample units and updating the confidence metric on the verification units; and performing training injection based on sample confidence and sample constraint relationships to obtain the updated detection model. This invention can perform unified calculations on real samples, weakly labeled samples, and synthetic samples, improving defect recall capability and suppressing false alarm fluctuations under varying process conditions.

The application discloses a discrete element analysis method for multi-scale random mechanical characteristics of soil and rock mixture, and comprises the following steps: obtaining real morphological characteristics of rock blocks; establishing a discrete element model based on the real morphological characteristics; generating a random sample space configuration; correcting particle contact relations; sequentially performing initial mechanical balance and triaxial shear loading on the corrected sample, obtaining stress-strain whole-process data, and extracting mechanical parameters; and based on the mechanical parameters, using Bootstrap resampling and taking a set variation coefficient as a convergence basis, obtaining a minimum simulation number, and constructing a quantitative relation table. The application solves the problems of morphological distortion, uneven distribution, unreasonable contact and arbitrary statistics existing in traditional analysis methods, improves simulation accuracy and engineering applicability, and can directly provide reliable parameters for slope, roadbed and dam body design.