6 September 2018 in IEEE Transactions on Evolutionary Computation
Most evolutionary optimization algorithms assume that the evaluation of the objective and constraint functions is straightforward. In solving many real-world optimization problems, however, such objective functions may not exist. Instead, computationally expensive numerical simulations or costly physical experiments must be performed for fitness evaluations. In more extreme cases, only historical data are available for performing optimization and no new data can be generated during optimization. Solving evolutionary optimization problems driven by data collected in simulations, physical experiments, production processes, or daily life are termed data-driven evolutionary optimization. In this paper, we provide a taxonomy of different data driven evolutionary optimization problems, discuss main challenges in data-driven evolutionary optimization with respect to the nature and amount of data, and the availability of new data during optimization. Real-world application examples are given to illustrate different model management strategies for different categories of data-driven optimization problems.
31 January 2018 in IEEE Transactions on Cybernetics
Gaussian processes (GPs) are the most popular model used in surrogate-assisted evolutionary optimization of computationally expensive problems, mainly because GPs are able to measure the uncertainty of the estimated fitness values, based on which certain infill sampling criteria can be used to guide the search and update the surrogate model. However, the computation time for constructing GPs may become excessively long when the number of training samples increases, which makes it inappropriate to use them as surrogates in evolutionary optimization. To address this issue, this paper proposes to use ensembles as surrogates and infill criteria for model management in evolutionary optimization. A heterogeneous ensemble consisting of a least square support vector machine and two radial basis function networks is constructed to enhance the reliability of ensembles for uncertainty estimation. In addition to the original decision variables, a selected subset of the decision variables and a set of transformed variables are used as inputs of the heterogeneous ensemble to further promote the diversity of the ensemble. The proposed heterogeneous ensemble is compared with a GP and a homogeneous ensemble for infill sampling criteria in evolutionary multiobjective optimization. Experimental results demonstrate that the heterogeneous ensemble is competitive in performance compared with GPs and much more scalable in computational complexity to the increase in search dimension.
1 January 2021 in IEEE Transactions on Systems, Man, and Cybernetics: Systems
Gaussian processes (GPs) are widely used in surrogate-assisted evolutionary optimization of expensive problems mainly due to the ability to provide a confidence level of their outputs, making it possible to adopt principled surrogate management methods, such as the acquisition function used in the Bayesian optimization. Unfortunately, GPs become less practical for high-dimensional multiobjective and many-objective optimization as their computational complexity is cubic in the number of training samples. In this article, we propose a computationally efficient dropout neural network (EDN) to replace the Gaussian process and a new model management strategy to achieve a good balance between convergence and diversity for assisting evolutionary algorithms to solve high-dimensional multiobjective and many-objective expensive optimization problems. While the conventional dropout neural network needs to save a large number of network models during the training for calculating the confidence level, only one single network model is needed in the EDN to estimate the fitness and its confidence level by randomly ignoring neurons in both training and testing the neural network. Extensive experimental studies on benchmark problems with up to 100 decision variables and 20 objectives demonstrate that, compared to state of the art, the proposed algorithm is not only highly competitive in performance but also computationally more scalable to high-dimensional many-objective optimization problems. Finally, the proposed algorithm is validated on an operational optimization problem of crude oil distillation units, further confirming its capability of handling expensive problems given a limited computational budget.
27 April 2026 in IEEE Transactions on Antennas and Propagation
A hybrid active sampling framework (HASF) combined with Gaussian process regression (GPR) is presented to achieve crucial sampling and efficient prediction of radar cross section (RCS). Specifically, the proposed framework involves three sequential stages to realize the global exploration, the local refine search, and the supplemental sampling. In Stage-Ⅰ, a novel fast leave-one-out (FLOO) method significantly enhances the computational efficiency of conventional LOO. The distance-based acquisition function is designed as the selection criterion. The establishment of a relative error threshold facilitates the realization of adaptive key sampling for diverse curves, thereby avoiding redundancy. In Stage-Ⅱ, the amount of local minimum neighborhoods that need to be finely searched can be determined adaptively according to data characteristics. The acquisition function based on distance and variance is constructed. It is employed as the selection criterion in order to execute a refine search in the selected local minimum neighborhoods. The same strategy as Stage-Ⅰ is employed in Stage-Ⅲ. Compared to conventional predict methods, HASF achieves higher prediction accuracy by the comprehensive sampling of both global and local regions. In comparison with other active sampling methodologies, HASF-FLOO has superior computational efficiency. The proposed framework has also demonstrated its effectiveness on complex models and measured datasets.