A Least Squares Ensemble Model Based on Regularization and Augmentation Strategy
Abstract
:1. Introduction
2. Background of Ensemble Methods
3. Proposed Regularized Least Squares Ensemble Model
3.1. Basic Formulation of the Least Squares Method
3.2. Samples Adding by the Augmentation Strategy
Algorithm 1 Pseudo code of augmentation strategy for adding samples |
Input: X = [ x1, x2, …, xN]. 1: Set empty, S = X. 2: Obtain 3 × Nadd samples by LHS, put them in Xlhs. 3: For i = 1: 3Nadd do 4: Calculate the distance of all the members in Xlhs to the samples in S. 5: Move the sample with the largest distance from Xlhs to and S. 6: End for 7: Construct KRG, RBF by S, calculate the uncertainties with (13) at the sample set , storage the difference values in Pkr. 8: Sort Pkr from the largest to the least, choose the top Nadd values of the corresponding samples, and put them into Xadd. Output: Xadd. |
3.3. The Regularization Strategy in the Least Squares System
- Random sampling N samples, the Nadd samples are obtained by the augmentation strategy, and the actual function values Y = [y1, y2, …, yN+Nadd] are calculated by expensive simulations.
- Choose N samples to construct the KRG, RBF, and SVR surrogate models, as the prediction values of the KRG and RBF at the N samples are equal to the corresponding actual function values of yi, i = 1, 2, …, N; calculate of the SVR model at each of the N samples.
- Evaluate , , and for the KRG, RBF, and SVR surrogate models, where i = 1, 2, …, Nadd and Nadd is the number of adding samples. Construct the matrix and Y as in Equation (14).
- Calculate the inverse of the augmented matrix system for by Equation (15), and the standardized weight factors by Equation (16).
Algorithm 2 Search for the optimal regularization parameter λ* |
Begin: 1: A constant array is set for λ, and l = min = 1, r = max = q. 2: While , , go to step 3, else go to step 8. 3: Randomly divide the predicted values of the KRG, RBF, and SVR surrogate models of the N + Nadd samples into k (we use k = 5 in this paper) equal parts. 4: The matrix is made up by the predicted values of three individual surrogate models in the k − 1 group, and by singular value decomposition (SVD), which can be expressed as 5: After the SVD, is calculated for λl and λr by: 6: Calculate the weight factors with Equations (15) and (16) for and , separately, and construct the and with Equation (1). 7: Calculate the RMSE of and with Equation (20), and the current optimal λc values are calculated as: 8: The optimal λ* is equal to the λc after iteration, and it can be used to construct the RLS-EM. Output: λ*. |
4. Case Studies
4.1. Numerical Examples
4.2. Deformation Prediction for the CNC Milling Machine Bed
5. Results
Author Contributions
Funding
Acknowledgments
Conflicts of Interest
Appendix A
Appendix A.1. Kriging (KRG)
Appendix A.2. Radial Basis Function (RBF)
Appendix A.3. Support Vector Regression (SVR)
Appendix B
Appendix B.1. Branin-Hoo Function
Appendix B.2. Camelback Function
Appendix B.3. Hartman Functions
aij | pij | ||||
---|---|---|---|---|---|
3.0 | 10 | 30 | 0.3689 | 0.1170 | 0.2673 |
0.1 | 10 | 35 | 0.4699 | 0.4387 | 0.7470 |
3.0 | 10 | 30 | 0.1091 | 0.8732 | 0.5547 |
0.1 | 10 | 35 | 0.03815 | 0.5743 | 0.8828 |
aij | pij | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
10 | 3.0 | 17.0 | 3.5 | 1.7 | 8.0 | 0.1312 | 0.1696 | 0.5569 | 0.0124 | 0.8283 | 0.5886 |
0.05 | 10.0 | 17.0 | 0.1 | 8.0 | 14.0 | 0.2329 | 0.4135 | 0.8307 | 0.3736 | 0.1004 | 0.9991 |
3.0 | 3.5 | 1.7 | 10.0 | 17.0 | 8.0 | 0.2348 | 0.1451 | 0.3522 | 0.2883 | 0.3047 | 0.6650 |
17.0 | 8.0 | 0.05 | 10.0 | 0.1 | 14.0 | 0.4047 | 0.8828 | 0.8732 | 0.5743 | 0.1091 | 0.0381 |
Appendix B.4. Extended-Rosenbrock Function
Appendix B.5. Dixion-Price Function
References
- Foorrester, A.; Keane, A. Recent advances in surrogate-based optimization. Prog. Aerosp. Sci. 2009, 45, 50–79. [Google Scholar] [CrossRef] [Green Version]
- Wang, G.G. Adaptive response surface method using inherited Latin hypercube design points. J. Mech. Des. 2003, 125, 210–220. [Google Scholar] [CrossRef]
- Nima, P.; Tang, X.W.; Yang, Q. Energy Evaluation of Triggering Soil Liquefaction Based on the Response Surface Method. Appl. Sci. 2019, 9, 694. [Google Scholar]
- Jerome, S.; Schiller, S.B. Designs for Computer Experiments. Technometrics 1989, 4, 41–47. [Google Scholar]
- Martin, J.D.; Simpson, T.W. Use of kriging models to approximate deterministic computer models. AIAA J. 2005, 43, 853–863. [Google Scholar] [CrossRef]
- Hwang, Y.; Cha, S.L.; Kim, S.; Jin, S.S.; Jung, H.J. The multiple-update-infill sampling method using minimum energy design for sequential surrogate modeling. Appl. Sci. 2018, 8, 481. [Google Scholar] [CrossRef]
- Fang, H.; Horstemeyer, M.F. Global response approximation with radial basis functions. Eng. Optim. 2006, 38, 407–424. [Google Scholar] [CrossRef]
- Zhang, J.H.; Chu, W.L.; Zhang, J.H.; Lv, Y. Vibroacoustic Optimization Study for the Volute Casing of a Centrifugal Fan. Appl. Sci. 2019, 9, 859. [Google Scholar] [CrossRef]
- Brereton, R.G.; Lloyd, G.R. Support vector machines for classification and regression. Analyst 2010, 135, 230–267. [Google Scholar] [CrossRef]
- Smola, A.J.; Schölkopf, B. A tutorial on support vector regression. Stat. Comput. 2004, 14, 199–222. [Google Scholar] [CrossRef] [Green Version]
- Liu, Z.Y.; Xu, Y.K.; Qiu, C.; Tan, J.R. A novel support vector regression algorithm incorporated with prior knowledge and error compensation for small datasets. Neural Comput. Appl. 2019. [Google Scholar] [CrossRef]
- Li, L.; Zheng, W.Z.; Wang, Y. Prediction of Moment Redistribution in Statically Indeterminate Reinforced Concrete Structures Using Artificial Neural Network and Support Vector Regression. Appl. Sci. 2019, 9, 28. [Google Scholar] [CrossRef]
- Jiang, H.; Zhang, Y.; Muljadi, E.; Zhang, J.J.; Gao, D.W. A short-term and high-resolution distribution system load forecasting approach using support vector regression with hybrid parameters optimization. IEEE Trans. Smart Grid 2018, 9, 3341–3350. [Google Scholar] [CrossRef]
- Wang, G.G.; Shan, S. Review of metamodeling techniques in support of engineering design optimization. J. Mech. Des. 2007, 4, 370–380. [Google Scholar] [CrossRef]
- Roy, R.; Hinduja, S.; Tet, R. Recent advances in engineering design optimisation: Challenges and future trends. CIRP Ann. 2008, 57, 697–715. [Google Scholar] [CrossRef]
- Bhosekar, A.; Ierapetritou, M. Advances in surrogate based modeling, feasibility analysis, and optimization: A review. Comput. Chem. Eng. 2018, 108, 250–267. [Google Scholar] [CrossRef]
- Liu, H.; Ong, Y.S.; Cai, J. A survey of adaptive sampling for global metamodeling in support of simulation-based complex engineering design. Struct. Multidiscip. Optim. 2018, 57, 393–416. [Google Scholar] [CrossRef]
- Song, X.; Lv, L.; Li, J.; Sun, W.; Zhang, J. An Advanced and Robust Ensemble Surrogate Model: Extended Adaptive Hybrid Functions. J. Mech. Des. 2018, 140, 041402. [Google Scholar] [CrossRef]
- Zhou, X.; Jiang, T. Metamodel selection based on stepwise regression. Struct. Multidiscip. Optim. 2016, 54, 641–657. [Google Scholar] [CrossRef]
- Wang, W.; Pei, J.; Yuan, S.; Zhang, J.; Yuan, J.; Xu, C. Application of different surrogate models on the optimization of centrifugal pump. J. Mech. Sci. Technol. 2016, 30, 567–574. [Google Scholar] [CrossRef]
- Song, X.; Sun, G.; Li, G.; Gao, W.; Li, Q. Crashworthiness optimization of foam-filled tapered thin-walled structure using multiple surrogate models. Struct. Multidiscip. Optim. 2013, 47, 221–231. [Google Scholar] [CrossRef]
- De Oliveira, M.A.; Possamai, O.; Dalla Valentina, L.V.; Flesch, C.A. Modeling the leadership-project performance relation: Radial basis function, Gaussian and Kriging methods as alternatives to linear regression. Expert Syst. Appl. 2013, 40, 272–280. [Google Scholar] [CrossRef]
- Worden, K.; Cross, E.J. On switching response surface models, with applications to the structural health monitoring of bridges. Mech. Syst. Signal Process. 2018, 98, 139–156. [Google Scholar] [CrossRef]
- Yang, H.; Xu, X.; Neumann, I. Optimal finite element model with response surface methodology for concrete structures based on Terrestrial Laser Scanning technology. Compos. Struct. 2018, 183, 2–6. [Google Scholar] [CrossRef]
- Kleijnen, J.P. Regression and Kriging metamodels with their experimental designs in simulation: A review. Eur. J. Oper. Res. 2017, 256, 1–16. [Google Scholar] [CrossRef] [Green Version]
- Mehdad, E.; Kleijnen, J.P. Stochastic intrinsic Kriging for simulation metamodeling. Appl. Stoch. Models Bus. Ind. 2018, 34, 322–337. [Google Scholar] [CrossRef]
- Zhang, Z.; Ou, J.P.; Li, D.S.; Zhang, S.F. Optimization Design of Coupling Beam Metal Damper in Shear Wall Structures. Appl. Sci. 2017, 7, 137. [Google Scholar] [CrossRef]
- Zhou, Q.; Wang, Y.; Jiang, P.; Shao, X.; Choi, S.K.; Hu, J.; Cao, L.; Meng, X. An active learning radial basis function modeling method based on self-organization maps for simulation-based design problems. Knowl.-Based Syst. 2017, 131, 10–27. [Google Scholar] [CrossRef]
- Fang, Y.; Zhan, Z.; Yang, J.; Liu, X. A Mixed-Kernel-Based Support Vector Regression Model for Automotive Body Design Optimization Under Uncertainty. ASCE-ASME J. Risk Uncertain. Eng. Syst. Part B Mech. Eng. 2017, 3, 041008. [Google Scholar] [CrossRef]
- Qi, J.; Hu, J. Multivariable case-based reason adaptation based on multiple-output support vector regression with similarity-related weight for parametric mechanical design. Adv. Mech. Eng. 2018, 10. [Google Scholar] [CrossRef]
- Goel, T.; Haftka, R.T.; Shyy, W.; Queipo, N.V. Ensemble of surrogates. Struct. Multidiscip. Optim. 2007, 33, 199–216. [Google Scholar] [CrossRef]
- Acar, E.; Rais-Rohani, M. Ensemble of metamodels with optimized weight factors. Struct. Multidiscip. Optim. 2009, 37, 279–294. [Google Scholar] [CrossRef]
- Zerpa, L.E.; Queipo, N.V.; Pintos, S.; Salager, J.L. An optimization methodology of alkaline-surfactant-polymer flooding processes using field scale numerical simulation and multiple surrogates. J. Pet. Sci. Eng. 2005, 47, 197–208. [Google Scholar] [CrossRef]
- Zhou, X.J.; Ma, Y.Z.; Li, X.F. Ensemble of surrogates with recursive arithmetic average. Struct. Multidiscip. Optim. 2011, 44, 651–671. [Google Scholar] [CrossRef]
- Acar, E. Various approaches for constructing an ensemble of metamodels using local measures. Struct. Multidiscip. Optim. 2010, 42, 879–896. [Google Scholar] [CrossRef]
- Lee, Y.; Choi, D.H. Pointwise ensemble of meta-models using v nearest points cross-validation. Struct. Multidiscip. Optim. 2014, 50, 383–394. [Google Scholar] [CrossRef]
- Ye, P.; Pan, G.; Dong, Z. Ensemble of surrogate based global optimization methods using hierarchical design space reduction. Struct. Multidiscip. Optim. 2018, 58, 537–554. [Google Scholar] [CrossRef]
- Liu, H.; Xu, S.; Wang, X.; Meng, J.; Yang, S. Optimal weighted pointwise ensemble of radial basis functions with different basis functions. AIAA J. 2016, 20, 3117–3133. [Google Scholar] [CrossRef]
- Gu, J.; Li, G.Y.; Dong, Z. Hybrid and adaptive meta-model-based global optimization. Eng. Optim. 2012, 44, 87–104. [Google Scholar] [CrossRef]
- Dong, H.; Song, B.; Wang, P.; Dong, Z. Hybrid surrogate-based optimization using space reduction (HSOSR) for expensive black-box functions. Appl. Soft Comput. 2018, 64, 641–655. [Google Scholar] [CrossRef]
- Ferreira, W.G.; Serpa, A.L. Ensemble of metamodels: The augmented least squares approach. Struct. Multidiscip. Optim. 2016, 53, 1019–1046. [Google Scholar] [CrossRef]
- Chen, L.; Qiu, H.; Jiang, C.; Cai, X.; Gao, L. Ensemble of surrogates with hybrid method using global and local measures for engineering design. Struct. Multidiscip. Optim. 2018, 57, 1711–1729. [Google Scholar] [CrossRef]
- Viana, F.A.; Haftka, R.T.; Steffen, V. Multiple surrogates: How cross-validation errors can help us to obtain the best predictor. Struct. Multidiscip. Optim. 2009, 39, 439–457. [Google Scholar] [CrossRef]
- Yin, H.; Wen, G.; Fang, H.; Qing, Q.; Kong, X.; Xiao, J.; Liu, Z. Multiobjective crashworthiness optimization design of functionally graded foam-filled tapered tube based on dynamic ensemble metamodel. Mater. Des. 2014, 55, 747–757. [Google Scholar] [CrossRef]
- Audet, C.; Kokkolaras, M.; Le Digabel, S.; Talgorn, B. Order-based error for managing ensembles of surrogates in mesh adaptive direct search. J. Glob. Optim. 2018, 70, 645–675. [Google Scholar] [CrossRef]
- Wang, H.; Jin, Y.; Sun, C.; Doherty, J. Offline data-driven evolutionary optimization using selective surrogate ensembles. IEEE Trans. Evolut. Comput. 2019, 23, 203–216. [Google Scholar] [CrossRef]
- Cattaneo, M.D.; Jansson, M.; Newey, W.K. Inference in linear regression models with many covariates and heteroscedasticity. J. Am. Stat. Assoc. 2018, 113, 1350–1361. [Google Scholar] [CrossRef]
- Lophaven, S.; Nielsen, H.; Sondergaard, J. DACE: A MATLAB Kriging Toolbox; Technical Report; Technical University of Denmark: Lyngby, Denmark, 2002. [Google Scholar]
- Sarra, S.A. The Matlab radial basis function toolbox. J. Open Res. Softw. 2017, 5, 8. [Google Scholar] [CrossRef]
- Chang, C.C.; Lin, C.J. LIBSVM: A library for support vector machines. ACM Trans. Intell. Syst. Technol. 2011, 2, 27. [Google Scholar] [CrossRef]
- Viana, F. SURROGATES Toolbox User’s Guide, version 3.0; SURROGATES Toolbox: Gainesville, FL, USA, 2011. Available online: https://sites.google.com/site/srgtstoolbox/ (accessed on 28 April 2019).
- Yao, P.; Xue, J.; Zhou, K. Study on the wire feed speed prediction of double-wire-pulsed MIG welding based on support vector machine regression. Int. J. Adv. Manuf. Technol. 2015, 79, 2107–2116. [Google Scholar] [CrossRef]
Function | ndv1 | N | Nadd | Nt |
---|---|---|---|---|
Branin-Hoo | 2 | 20 | 6 | 400 |
Camelback | 2 | 20 | 6 | 400 |
Hartman-3 | 3 | 30 | 9 | 1000 |
Hartman-6 | 6 | 100 | 30 | 1000 |
Extended Rosenbrock | 9 | 150 | 45 | 1000 |
Dixon–Price | 10 | 200 | 60 | 1000 |
ndv | Model | Details 1 |
---|---|---|
2 | KRG RBF SVR | Constant regression, Gaussian correlation, θ0 = ndv (1/2), 0.01 < θi < 20 Gaussian basis functions, kernel parameter γ = 4, no polynomial term Gaussian kernel γ = 5, regularization parameter C = ∞, quadratic loss ε = 0.01 |
3 | KRG RBF SVR | Constant regression, Gaussian correlation, θ0 = ndv (1/3), 0.01 < θi < 20 Gaussian basis functions, kernel parameter γ = 0.5, No polynomial term Gaussian kernel γ = 0.5, regularization parameter C = 100, quadratic loss ε = 0.01 |
6 | KRG RBF SVR | Linear regression, Gaussian correlation, θ0 = ndv (1/6), 0.01 < θi < 20 Gaussian basis functions, kernel parameter γ = 0.5, no polynomial term Gaussian kernel γ = 0.5, regularization parameter C = ∞, quadratic loss ε = 0.001 |
9 | KRG RBF SVR | Linear regression, Gaussian correlation, θ0 = ndv (1/9), 0.01 < θi < 20 Gaussian basis functions, kernel parameter γ = 1, polynomial term = 1 Gaussian kernel γ = 0.5, regularization parameter C = 100, quadratic loss ε = 0.001 |
12 | KRG RBF SVR | Quadratic regression, Gaussian correlation, θ0 = ndv(1/12), 0.01 < θi < 20 Gaussian basis functions, kernel parameter γ = 2, polynomial term = 1 Gaussian kernel γ = 0.5, regularization parameter C = 100, quadratic loss ε = 0.0001 |
Function | Metric | KRG | RBF | SVR | BP | PWS | NPWS | OWS | RLS-EM |
---|---|---|---|---|---|---|---|---|---|
Branin-Hoo | RMSE 1 AAE R2 | 11.855/4.452 6.032/1.971 0.941/0.045 | 19.127/4.461 11.278/2.020 0.859/0.068 | 18.981/4.591 11.315/2.154 0.860/0.071 | 13.869/5.920 7.577/3.312 0.917/0.068 | 15.451/4.372 8.670/1.942 0.905/0.055 | 15.582/4.333 8.771/1.902 0.904/0.055 | 15.184/4.475 8.460/2.063 0.908/0.056 | 12.008/4.679 6.436/2.317 0.939/0.050 |
Camel back | RMSE AAE R2 | 19.490/4.823 11.367/2.764 0.698/0.157 | 7.855/5.755 4.591/2.366 0.930/0.197 | 13.005/4.262 7.136/2.147 0.859/0.091 | 10.961/7.590 6.389/3.683 0.867/0.221 | 11.190/3.312 6.519/1.676 0.898/0.073 | 11.441/3.027 6.651/1.551 0.895/0.064 | 10.842/4.059 6.329/2.001 0.899/0.097 | 7.812/3.862 4.834/2.148 0.943/0.062 |
Hart mann-3 | RMSE AAE R2 | 0.235/0.060 0.159/0.036 0.929/0.038 | 0.417/0.049 0.281/0.030 0.788/0.049 | 0.372/0.076 0.221/0.035 0.826/0.077 | 0.253/0.083 0.170/0.051 0.914/0.061 | 0.273/0.045 0.176/0.024 0.907/0.032 | 0.278/0.044 0.179/0.023 0.905/0.032 | 0.265/0.048 0.171/0.027 0.913/0.033 | 0.233/0.044 0.157/0.034 0.931/0.030 |
Hart mann-6 | RMSE AAE R2 | 0.239/0.034 0.156/0.022 0.588/0.114 | 0.192/0.019 0.115/0.008 0.739/0.046 | 0.214/0.025 0.115/0.008 0.677/0.055 | 0.193/0.021 0.116/0.011 0.734/0.057 | 0.196/0.023 0.112/0.010 0.728/0.050 | 0.196/0.023 0.112/0.010 0.727/0.051 | 0.195/0.023 0.111/0.009 0.731/0.049 | 0.190/0.019 0.114/0.008 0.743/0.045 |
Extended Rosen-brock | RMSE (* 105) AAE (* 105) R2 | 0.201/0.023 0.154/0.017 0.765/0.028 | 0.185/0.021 0.142/0.015 0.801/0.024 | 0.201/0.027 0.154/0.019 0.764/0.036 | 0.187/0.022 0.144/0.017 0.796/0.027 | 0.180/0.021 0.138/0.016 0.811/0.023 | 0.181/0.022 0.138/0.016 0.811/0.023 | 0.180/0.022 0.138/0.016 0.812/0.022 | 0.184/0.020 0.141/0.015 0.808/0.023 |
Dixon–Price | RMSE (* 106) AAE (* 106) R2 | 0.159/0.017 0.238/0.034 0.835/0.025 | 0.195/0.022 0.151/0.017 0.753/0.034 | 0.230/0.029 0.174/0.022 0.760/0.045 | 0.161/0.019 0.151/0.019 0.831/0.033 | 0.175/0.020 0.166/0.020 0.802/0.026 | 0.177/0.020 0.168/0.020 0.797/0.026 | 0.171/0.019 0.162/0.019 0.810/0.025 | 0.158/0.019 0.160/0.019 0.841/0.057 |
Metric 1 | KRG | RBF | SVR | BP | PWS | NPWS | OWS | RLS-EM |
---|---|---|---|---|---|---|---|---|
RMSE | 87.86 | 82.15 | 115.53 | 82.97 | 81.88 | 88.61 | 89.82 | 79.87 |
AAE | 77.79 | 55.75 | 90.81 | 76.73 | 72.24 | 71.60 | 61.44 | 56.88 |
R2 | 0.82 | 0.82 | 0.85 | 0.82 | 0.84 | 0.86 | 0.85 | 0.87 |
© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Zhang, P.; Zhang, S.; Liu, X.; Qiu, L.; Yi, G. A Least Squares Ensemble Model Based on Regularization and Augmentation Strategy. Appl. Sci. 2019, 9, 1845. https://doi.org/10.3390/app9091845
Zhang P, Zhang S, Liu X, Qiu L, Yi G. A Least Squares Ensemble Model Based on Regularization and Augmentation Strategy. Applied Sciences. 2019; 9(9):1845. https://doi.org/10.3390/app9091845
Chicago/Turabian StyleZhang, Peng, Shuyou Zhang, Xiaojian Liu, Lemiao Qiu, and Guodong Yi. 2019. "A Least Squares Ensemble Model Based on Regularization and Augmentation Strategy" Applied Sciences 9, no. 9: 1845. https://doi.org/10.3390/app9091845
APA StyleZhang, P., Zhang, S., Liu, X., Qiu, L., & Yi, G. (2019). A Least Squares Ensemble Model Based on Regularization and Augmentation Strategy. Applied Sciences, 9(9), 1845. https://doi.org/10.3390/app9091845