Research On Optimization Based Onswarm Intelligence And Its Application

Nowadays many problems arising in economic, management and engineering fields can be seemed as optimization problems, such as facility location problem, vehicle routing problem, network flow design and so on. With the development of science and technology, real-world optimization problems become increasingly complex and are characterized by properties of high-dimensional, non-linear and large-scale. The practical complex problems impose new challenges on the research on optimization methods. The traditional methods for optimization problems include simplex algorithm, quadratic programming, Newton method, gradient method, etc. However, these methods have two main drawbacks: 1) specific mathematic requirements must be satisfied, e.g., convex, differentiable, and derivable; 2) the search capabilities for solving the complex and large-scale problems are lacking. As a result, more efficient methods are always in need.Swarm intelligence is a novel bio-inspired heuristic method. Since its introduction swarm intelligence has attracted many interests due to its advantages including simple, intelligent and efficient features. Swarm intelligence has been applied to deal with many optimization problems. In this study, in view of different perspectives of bio-simulation and various search behaviors, particle swarm optimization(PSO) and bacterial foraging optimization(BFO) are adopted as representatives of swarm intelligence for further research. The research goal is to conduct specific improvement on swarm intelligence algorithms to overcome current research gaps. The main contributions are summarized as the followings:(1) The standard PSO has several main criticisms: 1) the utilization of historical information is less than satisfactory; 2) the loss of population diversity is serious especially during the late evolution; 3) the underperforming individuals also provide misleading information and consume computation cost. To address these problems, an orthogonal-hybrid learning particle swarm optimization is proposed. A permutation strategy based on orthogonal experimental design is developed as a metabolic mechanism to enhance population diversity. In addition, a hybrid learning strategy is designed to exploit the particles’ best experiences and guide individuals more efficiently. Experimental study justify the effectiveness and efficiency of the proposed method.(2) The standard PSO’s performance on complex problems is still unsatisfactory. Although researchers proposed multiple-swarm PSO to avoid premature on complex problems, three main shortcomings still exists. First, most MS-PSOs were developed for particular problems so their search capability on diverse landscapes is lacking. Moreover, research on MS-PSO has so far treated the sub-swarms as cooperative groups with minimum competition(if not none). In addition, the size of each sub-swarm is set to be fixed which may encounter excessive computation cost. In order to solve these issues, a novel optimizer using adaptive heterogeneous particle swarms(AHPS2) is developed. In the AHPS2, multiple heterogeneous swarms, each consisting of a group of homogenous particles having similar learning strategies, are introduced. An adaptive competition strategy is proposed so the size of each swarm can be dynamically adjusted based on its group performance. Besides, two complementary search techniques and two population adjustment models are analyzed. Simulation results indicate that the proposed strategies improve PSO’s performance on diverse landscapes.(3) The disadvantages of the standard BFO include: 1) low solution accuracy; 2) slow convergence; 3) performance deteriorates seriously with increase of dimensionality. To address these issues, a global cooperative BFO is proposed. In this method, a global cooperative search model is built to alter the movement updating of bacteria. To improve accuracy and speed of BFO, the original chemotaxis is modified by the information sharing and cooperative search. By learning from the superior individuals for dimensionally updating, the strategy enables all bacteria except the worst one working as potential exemplars to collaborate with others to refine the better solutions. Besides, two models, i.e., with/without swimming behavior, are studied. Experimental results demonstrate that the proposed method improve BFO’s performance on different problems with various dimensions.(4) The original BFO is characterized by strong exploitation, easily escaping from local optima and weak global convergence. PSO is good at global convergence, but bad at local exploitation and getting rid of local optima. This research goal is to take advantage of BFO and PSO for an enhanced swarm intelligence by building a hybrid framework of swarm intelligence. Within the framework, the memory scheme of historical information from individual and population in PSO is employed to decide next movement; the stochastic swimming scheme in BFO is adopted to improve local search capability. Besides, PSO is in charge of global exploration, while BFO is taking over the local exploitation. Simulation experiments show that the novel hybrid framework of swarm intelligence better takes advantage of the strengths of BFO and PSO and outperforms any one of them during experiments.(5) To justify the effectiveness and efficiency of the proposed approaches for real-world problems, the methods are applied to solve two engineering problems in economic and management fields. First, the capacitated distribution center location problem is modeled. The solution methodologies based on the improved algorithms are presented to optimized the model. Simulation results show that the proposed methods are able to reduce the system cost when compared with other algorithms. Second, the constrained cardinality portfolio optimization problem is analyzed and modeled. The enhanced swarm intelligence algorithms are employed to address the problem. Simulation experiments show that the proposed methods approximate the efficient frontier that is close to the standard efficient frontier. It can be concluded the developed approaches’ efficiency on optimizing constrained cardinality portfolio optimiztion problem.