Guang Pan

Bio: Guang Pan is an academic researcher. The author has contributed to research in topics: Ensemble forecasting. The author has an hindex of 1, co-authored 1 publications receiving 1 citations.

Selecting the Best Quantity and Variety of Surrogates for an Ensemble Model

Abstract: Surrogate modeling techniques are widely used to replace the computationally expensive black-box functions in engineering. As a combination of individual surrogate models, an ensemble of surrogates is preferred due to its strong robustness. However, how to select the best quantity and variety of surrogates for an ensemble has always been a challenging task. In this work, five popular surrogate modeling techniques including polynomial response surface (PRS), radial basis functions (RBF), kriging (KRG), Gaussian process (GP) and linear shepard (SHEP) are considered as the basic surrogate models, resulting in twenty-six ensemble models by using a previously presented weights selection method. The best ensemble model is expected to be found by comparative studies on prediction accuracy and robustness. By testing eight mathematical problems and two engineering examples, we found that: (1) in general, using as many accurate surrogates as possible to construct ensemble models will improve the prediction performance and (2) ensemble models can be used as an insurance rather than offering significant improvements. Moreover, the ensemble of three surrogates PRS, RBF and KRG is preferred based on the prediction performance. The results provide engineering practitioners with guidance on the superior choice of the quantity and variety of surrogates for an ensemble.

Integration of Second-Order Sensitivity Method and CoKriging Surrogate Model

Abstract: The global exploring feature of the surrogate model makes it a useful intermedia for design optimization. The accuracy of the surrogate model is closely related with the efficiency of optima-search. The cokriging approach described in present studies can significantly improve the surrogate model accuracy and cut down the turnaround time spent on the modeling process. Compared to the universal Kriging method, the cokriging method interpolates not only the sampling data, but also on their associated derivatives. However, the derivatives, especially high order ones, are too computationally costly to be easily affordable, forming a bottleneck for the application of derivative enhanced methods. Based on the sensitivity analysis of Navier–Stokes equations, current study introduces a low-cost method to compute the high-order derivatives, making high order derivatives enhanced cokriging modeling practically achievable. For a methodological illustration, second-order derivatives of regression model and correlation models are proposed. A second-order derivative enhanced cokriging model-based optimization tool was developed and tested on the optimal design of an automotive engine cooling fan. This approach improves the modern optimal design efficiency and proposes a novel direction for the large scale optimization problems.

The use of Fuzzy rule-based systems in the design process of the metallic products on example of microstructure evolution prediction

Abstract: Abstract The challenge on the contemporary market of consumer goods is a quick response to customer needs. It entails time restrictions, which a semi-finished products’ (including metal products) manufacturer must meet. This issue must be addressed during a design phase, which for the most of semi-finished products suppliers, takes part during a quotation preparation process. Our research is aimed at investigating possibility of application of Fuzzy Reasoning methods for shortening of a design process, being a part of this process. We present a study on application of simplified models for solving technological tasks, allowing obtaining expected properties of designed products. The core of our concept is replacing numerical models and classical metamodels with a rule-based reasoning. A quotation preparation process can be supported by solving a technological problem without numerical experiments. Our goal was to validate the thesis basing not only on the presentation of some potential solutions but also on the results of simulation studies. The problem is illustrated with an example of thermal treatment of aluminum alloys, aimed at evaluation of a summary fraction of precipitations as a function of time and technological parameters. We assumed that it is possible to use both unstructured and point numerical experiments for knowledge acquisition. Implementation of this concept required the use of hybrid knowledge acquisition methods that combine the results of point experiments with expert knowledge. A comparison of obtained results to the ones obtained with metamodels shows a similar efficiency of both approaches, while our method is less time and laborious.