Neural Network Surrogate Model Ensemble for Sparse Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-fidelity simulation tools like FEA and CFD are computationally expensive, limiting the number of simulations that can be conducted for optimal design, and existing surrogate models perform poorly with sparse data, leading to suboptimal designs and high prediction errors.
Innovation Solution
The development of neural-network based surrogate models that create a pool of neural networks, generate ensembles using multi-objective functions, select local ensembles, and combine them to form a global ensemble, using a fidelity, complexity, and ambiguity evolutionary selection (FCAES) algorithm to improve prediction robustness and generalization in sparse data conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-fidelity simulation tools (FEA, CFD) are used for design optimization, then prediction accuracy is improved, but computational cost increases
Solution Approach 1:
The patent creates surrogate models that are simplified copies of high-fidelity simulation tools. These surrogate models replicate the input-output behavior of complex FEA/CFD simulations but with significantly reduced computational cost, enabling rapid design optimization without requiring expensive repeated simulations.
Solution Approach 2:
The patent employs neural network ensembles as computationally inexpensive alternatives to expensive simulation tools. Once trained on a limited set of high-fidelity simulation data, these neural network models can be executed rapidly multiple times for design optimization, replacing the need for repeated expensive simulations.
2Loss of energy
If existing surrogate models are used with sparse data, then computational cost is reduced, but prediction error increases
Solution Approach 1:
The patent combines multiple neural networks into ensembles to improve prediction accuracy with sparse data. By merging the predictions of multiple individual neural networks, the ensemble approach reduces prediction variance and improves generalization performance compared to single neural networks, while still maintaining low computational cost.
Solution Approach 2:
The patent creates composite predictive models by combining multiple neural networks with different architectures, training data, or initialization into ensembles. This composite approach leverages the strengths of individual networks and compensates for their weaknesses, achieving superior prediction accuracy with limited training data.
Data Source
AI summary
Various neural-network based surrogate model construction methods are disclosed herein, along with various applications of such models. Designed for use when only a sparse amount of data is available (a “sparse data condition”), some embodiments of the disclosed systems and methods: create a pool of neural networks trained on a first portion of a sparse data set; generate for each of various multi-objective functions a set of neural network ensembles that minimize the multi-objective function; select a local ensemble from each set of ensembles based on data not included in said first portion of said sparse data set; and combine a subset of the local ensembles to form a global ensemble. This approach enables usage of larger candidate pools, multi-stage validation, and a comprehensive performance measure that provides more robust predictions in the voids of parameter space.