Multi-fidelity global optimization using a data-mining strategy for computationally intensive black-box problems

2021 ◽  
pp. 107212
Author(s):  
Jie Liu ◽  
Huachao Dong ◽  
Peng Wang
Keyword(s):  
Data Mining ◽  
Global Optimization ◽  
Black Box ◽  
Computationally Intensive ◽  
Data Mining Strategy

Redesign of casing treatment for a transonic centrifugal compressor based on a hybrid global optimization method

2021 ◽  
pp. 095440622110398 ◽  
Author(s):  
Xiaojian Li ◽  
Zhengxian Liu ◽  
Yijia Zhao
Keyword(s):  
Global Optimization ◽  
Surrogate Model ◽  
Flow Characteristics ◽  
Optimal Solution ◽  
Black Box ◽  
Expected Improvement ◽  
Local Flow ◽  
Casing Treatment ◽  
Highly Nonlinear ◽  
Computationally Intensive

As a typical black-box problem, recirculating casing treatment (RCT) optimization of compressor stages is computationally intensive and time consuming, even though surrogate models are usually employed. In order to improve efficiency and robustness of the optimization, an expected-improvement (EI) based hybrid global optimization (EHGO) algorithm is developed by coupling an EI-based surrogate model with a hybrid optimization algorithm. Highly nonlinear and multiple modality mathematical tests show that the EHGO algorithm is able to create a high-fidelity surrogate model near targeted regions with less evaluated samples, and to obtain the global optimal solution simultaneously. The RCT of a compressor stage is optimized based on this algorithm. The number of CFD simulations required for obtaining an optimum solution is greatly reduced, as compared to similar studies using conventional methods. The optimization results show that the aerodynamic performance is improved over the whole speed line and the flow range is also extended. The dominant factors for the performance improvements and the enhanced stall margin are addressed by analyzing the local flow characteristics before and after optimization. It is found that those factors include: removing a larger amount of low-momentum fluid, achieving a more uniform flow of impeller passage in circumferential direction, and reducing the radial distortion of impeller inlet flow. The proposed algorithm has the potential to considerably speed up the optimization process and make the optimization much more accessible. It can be generalized to deal with other computationally intensive black-box problems, for example, turbomachinery optimization.

scholarly journals A method for convex black-box integer global optimization

2021 ◽  
Author(s):  
Jeffrey Larson ◽  
Sven Leyffer ◽  
Prashant Palkar ◽  
Stefan M. Wild
Keyword(s):  
Global Optimization ◽  
Black Box

scholarly journals Data-driven analysis of interactions between people with dementia and a tablet device

2017 ◽  
Vol 3 (2) ◽  
pp. 735-738
Author(s):  
Wolfgang Doneit ◽  
Jana Lohse ◽  
Kristina Glesing ◽  
Clarissa Simon ◽  
Monika Fischer ◽  
...  
Keyword(s):  
Data Mining ◽  
Facial Expressions ◽  
Music Videos ◽  
Data Driven ◽  
Tablet Devices ◽  
Log Files ◽  
People With Dementia ◽  
Analysis Strategy ◽  
Tablet Device ◽  
Data Mining Strategy

AbstractIn the project I-CARE a technical system for tablet devices is developed that captures the personal needs and skills of people with dementia. The system provides activation content such as music videos, biographical photographs and quizzes on various topics of interest to people with dementia, their families and professional caregivers. To adapt the system, the activation content is adjusted to the daily condition of individual users. For this purpose, emotions are automatically detected through facial expressions, motion, and voice. The daily interactions of the users with the tablet devices are documented in log files which can be merged into an event list. In this paper, we propose an advanced format for event lists and a data analysis strategy. A transformation scheme is developed in order to obtain datasets with features and time series for popular methods of data mining. The proposed methods are applied to analysing the interactions of people with dementia with the I-CARE tablet device. We show how the new format of event lists and the innovative transformation scheme can be used to compress the stored data, to identify groups of users, and to model changes of user behaviour. As the I-CARE user studies are still ongoing, simulated benchmark log files are applied to illustrate the data mining strategy. We discuss possible solutions to challenges that appear in the context of I-CARE and that are relevant to a broad range of applications.

Design Optimization of a 3D Parameterized Vane Cascade With Non-Axisymmetric Endwall Based on a Modified EGO Algorithm and Data Mining Techniques

2017 ◽  
Author(s):  
Chenxi Li ◽  
Zhendong Guo ◽  
Liming Song ◽  
Jun Li ◽  
Zhenping Feng
Keyword(s):  
Data Mining ◽  
Global Optimization ◽  
Design Space ◽  
Pressure Coefficient ◽  
Differential Evolution Algorithm ◽  
Optimal Solution ◽  
Efficient Global Optimization ◽  
Data Mining Techniques ◽  
Parallel Axis ◽  
Better Than

The design of turbomachinery cascades is a typical high dimensional and computationally expensive problem, a metamodel-based global optimization and data mining method is proposed to solve it. A modified Efficient Global Optimization (EGO) algorithm named Multi-Point Search based Efficient Global Optimization (MSEGO) is proposed, which is characterized by adding multiple samples at per iteration. By testing on typical mathematical functions, MSEGO outperforms EGO in accuracy and convergence rate. MSEGO is used for the optimization of a turbine vane with non-axisymmetric endwall contouring (NEC), the total pressure coefficient of the optimal vane is increased by 0.499%. Under the same settings, another two optimization processes are conducted by using the EGO and an Adaptive Range Differential Evolution algorithm (ARDE), respectively. The optimal solution of MSEGO is far better than EGO. While achieving similar optimal solutions, the cost of MSEGO is only 3% of ARDE. Further, data mining techniques are used to extract information of design space and analyze the influence of variables on design performance. Through the analysis of variance (ANOVA), the variables of section profile are found to have most significant effects on cascade loss performance. However, the NEC seems not so important through the ANOVA analysis. This is due to the fact the performance difference between different NEC designs is very small in our prescribed space. However, the designs with NEC are always much better than the reference design as shown by parallel axis, i.e., the NEC would significantly influence the cascade performance. Further, it indicates that the ensemble learning by combing results of ANOVA and parallel axis is very useful to gain full knowledge from the design space.

A Survey of Feature Selection Techniques

2011 ◽  
pp. 1888-1895 ◽  
Author(s):  
Barak Chizi ◽  
Lior Rokach ◽  
Oded Maimon
Keyword(s):  
Data Mining ◽  
Feature Selection ◽  
Mining Method ◽  
Data Set ◽  
Data Mining Method ◽  
Data Mining Algorithms ◽  
Wrapper Approach ◽  
Computationally Intensive ◽  
Filter Approach ◽  
Mining Algorithms

Dimensionality (i.e., the number of data set attributes or groups of attributes) constitutes a serious obstacle to the efficiency of most data mining algorithms (Maimon and Last, 2000). The main reason for this is that data mining algorithms are computationally intensive. This obstacle is sometimes known as the “curse of dimensionality” (Bellman, 1961). The objective of Feature Selection is to identify features in the data-set as important, and discard any other feature as irrelevant and redundant information. Since Feature Selection reduces the dimensionality of the data, data mining algorithms can be operated faster and more effectively by using Feature Selection. In some cases, as a result of feature selection, the performance of the data mining method can be improved. The reason for that is mainly a more compact, easily interpreted representation of the target concept. The filter approach (Kohavi , 1995; Kohavi and John ,1996) operates independently of the data mining method employed subsequently -- undesirable features are filtered out of the data before learning begins. These algorithms use heuristics based on general characteristics of the data to evaluate the merit of feature subsets. A sub-category of filter methods that will be refer to as rankers, are methods that employ some criterion to score each feature and provide a ranking. From this ordering, several feature subsets can be chosen by manually setting There are three main approaches for feature selection: wrapper, filter and embedded. The wrapper approach (Kohavi, 1995; Kohavi and John,1996), uses an inducer as a black box along with a statistical re-sampling technique such as cross-validation to select the best feature subset according to some predictive measure. The embedded approach (see for instance Guyon and Elisseeff, 2003) is similar to the wrapper approach in the sense that the features are specifically selected for a certain inducer, but it selects the features in the process of learning.

Sign in / Sign up

Export Citation Format

Share Document