Papers by Keyword: Differential Evolution

Abstract: This work minimises the operational costs of synthetic jet fuel production (C₈ – C₁₆ hydrocarbon range) by manipulating five operating parameters of the power-to-liquid (PtL) process. Conservative estimates for variable expenses (90^th percentiles of historical costs) and revenues (10^th percentiles of prices for revenues) were assumed for optimisation of the simulation in Aspen Plus^TM software. The distributions of costs under parametric uncertainty were obtained using the Monte Carlo algorithm. Two optimisation scenarios for cost minimisation were developed. They differ on the assumption that CO₂ emitted to air is taxed. An additional optimisation scenario targets the maximization of jet fuel production. In comparison to the non-optimised case, the optimised configurations reduce CO₂ emissions between 17 and 19% while decreasing the operating costs of jet fuel production by at least 0.80 – 1.89 € / L (depending on the scenario) with a stochastic probability of 90%, and a probability of 50% of reducing them by 0.93 – 2.09 € / L.

Abstract: In the present work it is investigated the performance of an algorithm that associates Differential Evolution and Constructal Design for geometrical optimization of a heat transfer problem. It is considered the intrusion of a cooled Double T-shaped cavity into a rectangular conducting solid wall with internal heat generation. The main purpose here is to evaluate the algorithm capability to reproduce the effect of geometric ratios over the dimensionless maximum excess of temperature (performance indicator of the heat transfer problem), as well as, the influence of Differential Evolution (DE) parameters over the optimization analysis. The definition of search space for each degree of freedom and problem constraints is performed with Constructal Design, while the Differential Evolution algorithm is used in the optimization process. Here parameters as mutation operator (M), crossover constant (CR), differential amplification factor (F), Population Size (PS) and Generations number (G) are evaluated. A theoretical recommendation about the suitable parameters set for the optimization algorithm for this kind of heat transfer problem is proposed. Results indicated that the crossover constant (CR) and amplification factor (F) are important parameters for suitable prediction of the effect of degrees of freedom over thermal performance. Moreover, when CR = 0.7 and F = 1.5 results obtained with the algorithm are more robust for the achievement of the best shapes and requires lower number of iterations (IT = PS × G) for reproduction of effect of geometric variables over performance.

Abstract: Permutation flowshop scheduling problem (PFSP) is a classical NP-hard combinatorial optimization problem, which provides a challenge for evolutionary algorithms.Since it has been shown that simple evolutionary algorithms cannot solve the PFSP efficiently, local search methods are often adopted to improve the exploitation ability of evolutionary algorithms. In this paper, a hybrid differential evolution algorithm is developed to solve this problem. This hybrid algorithm is designed by incorporating a dynamic variable neighborhood search (DVNS) into the differential evolution. In the DVNS, the neighborhood is based on multiple moves and its size can be dynamically changed from small to large so as to obtain a balance between exploitation and exploration. In addition, a population monitoring and adjusting mechanism is also incorporated to enhance the search diversity and avoid being trapped in local optimum.Experimental results on benchmark problems illustrated the efficiency of the proposed algorithm.

Abstract: Porosity is a major problem occurring in aluminium alloy casting. During the process of solidification, alloy would shrink and emit dissolving hydrogen causing porosity formation inside the solidified part which leads to mechanical properties degradation. This research aims to produce a formula to explain the resulting porosity with the initial chemical compositions and cooling rate. A mathematic model is, at first, inferred from previous researches to be a template function. Differential Evolution is utilized to generate inner polynomial parts and to find appropriate coefficients to experimental data obtained from other publications. The optimized function promisingly shows good fit to the problem domain demonstrating that the resulting function is an efficient model to explain porosity formation behaviour.

Abstract: Economic load dispatch is an important energy planning process. In order to maximize profit, proper analysis must be conducted by the utilities to reduce the operating cost. It includes the correct selection of fuel type with respect to the required output level as different fuel type will have different price signals. Usually, a piecewise quadratic function is used to represent the multiple fuel options. In this paper, fuzzy-based technique is proposed to select the suitable fuel type. The tests involve IEEE 57-Bus system with 7 generators. The economic load dispatch with a multi-fuel problem is also optimized by using DEIANT algorithms to compute the feasible operating cost and minimizing the power loss.

Abstract: A multi-strategy based population optimization, referred to MSPO, is proposed in this paper. The algorithm is developed by hybridizing four different population-based algorithms, bare bone particle swarm optimization, quantum-behaved particle swarm optimization, differential evolution and opposition-based learning. It aims at enhancing the exploration and exploitation capability of population based algorithm for general optimization problem. These four options are randomly selected with equal probability during the search process. The proposed algorithm is validated against test functions and then compares its performance with those of particle swarm optimization and bare bone particle swarm optimization. Numerical results show that the performance is increased greatly both in solution quality and convergent speed.

Abstract: Differential Evolution Algorithm (DE) is fast and stable, but it’s easy to fall into the local optimal solution and the population diversity reduces fast in the later period. In order to improve the algorithm optimization and convergence capability, this paper proposes an improved DE algorithm based on the new crossover strategy (CMDE). As to the Crossover-factor is decided by the proportion of the variance and the evolution process in each generation, so it can follow the process of evolution and constantly change; the added operation of Second Mutation can improve the capacity of solving problem, which algorithm falls into the local solution easily. With four standard test functions, the results show that the CMDE algorithm is superior to DE in convergence speed, precise and stability of algorithm.

Abstract: Hyperspectral image classification is difficult due to the high dimensional features but limited training samples. Tri-training learning is a widely used semi-supervised classification method that addresses the problem of lacking of labeled examples. In this paper, a novel semi-supervised learning algorithm based on tri-training method is proposed. The proposed algorithm combines margin sampling (MS) technique and differential evolution (DE) algorithm to select the most informative samples and perturb them randomly. Then the samples we obtained, which can fulfill the labeled data distribution and introduce diversity to multiple classifiers, are added to training set to train base classifiers for tri-training. The proposed algorithm is experimentally validated using real hyperspectral data sets, indicating that the combination of MS and DE can significantly reduce the need of labeled samples while achieving high accuracy compared with state-of-the-art algorithms.

Abstract: The multistage goal programming model is popular to model the defense projects portfolio optimization problem in recent years. However, as its high-dimensional variables and large-scale solution space, the addressed model is hard to be solved in an acceptable time. To deal with this challenge, we propose an improved differential evolution algorithm which combines three novel strategies i.e. the variables clustering based evolution, the whole randomized parameters, and the child-individual based selection. The simulation results show that this algorithm has the fastest convergence and the best global searching capability in 6 test instances with different scales of solution space, compared with classical differential evolution algorithm (CDE), genetic algorithm (GA) and particle swarm optimization (PSO) algorithm.

Abstract: This paper presents a differential evolution algorithm with designed greedy heuristic strategy to solve the task scheduling problem. The static task scheduling problem is NP-complete and is a critic issue in parallel and distributed computing environment. A vector consists of a task permutation assigned to each individual in the target population by using DE mutation and crossover operators. A heuristic strategy is used to generate the feasible solutions as there a lot of infeasible solutions in the solution space as the size of the problem increase. And the strategies of the particle swarm algorithm are employed to modify the DE crossover operator for speeding up the search to optimal solution. And then, the individual is replaced with the corresponding target individual if it is global best or local best in terms of fitness. The performance of the algorithm is illustrated by comparing with the existing effectively scheduling algorithms. The performances of the proposed algorithms are tested on the benchmark and compared to the best-known solutions available. The computational results demonstrate that effectively and efficiency of the presented algorithm.