This paper proposes a surrogate-assisted evolutionary algorithm to tackle expensive inequality-constrained optimization problems through global exploration and local exploitation. The algorithm begins with an exploration stage that involves sampling in three kinds of global regions: the feasible region, the better-objective region, and the converging region. Specifically, sampling in the uncertain feasible region mitigates issues caused by inaccurate objective surrogates. In addition, sampling in the uncertain region containing better objective values than the current best feasible solution reduces the risk of missing the global optimum due to inaccurate constraint surrogates. Moreover, sampling in the converging region facilitates quick convergence to the global feasible optimum. Following the exploration stage, promising feasible and infeasible solutions are further refined using local surrogate-based search strategies. To address the risk of missing the global optimum resulting from limited local region scope, the regions are adaptively extended if predicted infill points lie on the boundary. If an infill point is determined to showcase a better objective value after accurate evaluation, a rewarding local search is performed within the local region. This exploration-exploitation process iterates until the computation budget is exhausted. Experimental results demonstrate that the proposed algorithm outperforms the selected state-of-the-art algorithms on the majority of tested problems.

By decomposing the problem into a series of low-dimensional subproblems, cooperative coevolution is an effective method for large-scale optimization problems. This work reveals that when constraints are introduced in decomposition-based methods, the optima of a subproblem might change during the evolution process. Therefore, it is essential to maintain the population diversity in cooperative coevolution. This work proposes a bi-population enhanced cooperative differential evolution to address this issue. In the proposed method, the population of a subproblem is divided into two subpopulations (local and global) according to a specific strategy. The global and local subpopulations evolve independently, using different differential mutation operators to generate offspring separately without interference. The local subpopulation aims to track and improve the previous optima, while the global subpopulation attempts to find and locate the potential emerging optima. The proposed algorithm is tested on 12 constrained large-scale benchmarks and the experiments show that it can provide highly competitive performance compared to state-of-the-art algorithms. The proposed bi-population strategy is more effective at the lower dimensionality of the subproblem.

In this work, an adaptive constraint-handling approach is developed to improve the efficiency of surrogate-based optimization (SBO). Similar to other SBO methods, the proposed approach is a sequential updating process, whereas two candidate points considering the significance of objective and constraints are generated respectively in each cycle. In detail, the candidate point of objective is obtained through the penalized lower confidence bounding (PLCB) infill criterion. Additionally, an infill criterion of the constraints (called MLCB) which can accurately characterize the boundaries of the constraints is developed to determine the candidate point of constraints. Then, a selection algorithm is developed to select one or two candidate point(s) as the new training point(s) adaptively according to the current optimal value and the accuracy of the constraint boundaries. The selection algorithm is composed of three phases. In the first phase, the candidate point of constraints is selected to find a feasible solution. Two candidate points of objective and constraints are added to speed up the convergence in the second phase. In the third phase, the candidate point of objective is chosen to improve the quality of the feasible optimal solution. The proposed approach is tested on seven numerical functions and compared with state-of-the-art methods. Results indicate that the proposed approach has excellent global optimization ability, meanwhile, it reduces significantly computational resources.

Multi-objective optimization problems (MOPs) are commonly encountered in practical engineering. Multi-objective evolutionary algorithms (MOEAs) are one of the powerful methods to solve MOPs. However, MOEAs require a large number of fitness evaluations, which limits the practical application of MOEAs. Surrogate model assisted evolutionary algorithm (SAEA) can effectively alleviate the computation burden of MOEAs by replacing time-consuming simulation with the surrogate model. In this paper, a three-stage adaptive multifidelity surrogate (MFS) model assisted multi-objective genetic algorithm(MOGA) are proposed. In the first stage, a cheap lowfidelity (LF) model is adopted to obtain a preliminary Pareto frontier (PF). In the second stage, some of the individuals are selected and sent to high-fidelity (HF) model to construct MFS models, which are used to evaluate the fitness functions and sequentially updated according to the model management strategy. During this stage, in order to obtain a better PF, a fidelity control strategy is developed to subjectively determine when transforming is conducted to the third stage, in which all the individuals are evaluated by the HF model. Three benchmark tests are used to test the performance of the proposed method. Results show that the proposed method performs better than online MFS model assisted MOGA(OLMFM-MOGA) and NSGA-II with HF model, especially when the correlation between the LF and HF models is very poor.

The existing multiobjective expected improvement (EI) criteria are often computationally expensive because they are calculated using multivariate piecewise integrations, the number of which increases exponentially with the number of objectives. In order to solve this problem, this paper proposes a new approach to develop cheap-to-evaluate multiobjective EI criteria based on the proposed EI matrix (EIM). The elements in the EIM are the single-objective EIs that the studying point has beyond each Pareto front approximation point in each objective. Three multiobjective criteria are developed by combining the elements in the EIM into scalar functions in three different ways. These proposed multiobjective criteria are calculated using only 1-D integrations, whose number increases linearly with respect to the number of objectives. Moreover, all the three criteria are derived in closed form expressions, thus are significantly cheaper to evaluate than the state-of-the-art multiobjective criteria. The efficiencies of the proposed criteria are validated through 12 test problems. Besides the computational advantage, the proposed multiobjective EI criteria also show competitive abilities in approximating the Pareto fronts of the chosen test problems compared against the state-of-the-art multiobjective EI criteria.

Recently, more and more people are focused on the energy efficiency of HPC(High Performance Computing) centers. The power consumption of supercomputer listed in the first 10 of top 500 is usually more than 1Mkw. Power-aware techniques have been widely used in data centre or supercomputer centre for reducing the power consumption. It is useful for us to reduce the power consumption with frequency scalable CPU being adopted in the HPC centers. Because load unbalance is a common problem in high performance computing, we can gear the frequency of CPU down when it is idle so that the power consumption can be reduced with little performance lose. In this paper, we present a new method called AESC(Automatic Energy Status Controlling) which can control the energy status of CPU automatically. Firstly, we identify the application as different kinds of power consumption by the load balancing and communication time. Secondly, we will give a control policy for controlling the energy status based on former results. By using this policy, we can automatically gear the frequency of CPU and get the best result that 20% less energy while increasing execution time by only 1%.