The Improved And Applied Research Of Gravitational Search Algorithm

There are lots of optimization problems to be solved in the field of engineering,management science, information processing, manufacturing and many other fields. Currentlythe optimization algorithms used in dealing with these optimization problems are mainlytraditional optimization methods and intelligent optimization algorithms. The intelligentoptimization algorithm is a stochastic optimization algorithm by simulating the behavior ofnatural biological communities, and extracting the model from their group behavior.Compared with the traditional optimization algorithm, swarm intelligence optimizationalgorithm has higher adaptability, robustness, parallel processing capability and many otheradvantages. Massive applications demonstrate that the swarm intelligence algorithm is a goodtool to solve complex single-objective and multi-objective optimization problems.Gravitational search algorithm is a new swarm intelligence algorithm, and its basic ideais based on Newton’s law of universal gravitation. Gravitational search algorithm hasachieved good application in some areas and attracted many domestic and foreign scholars tostudy it because of its simple principle, wide universality and its high efficiency in solving thenonlinear functions. This article will study the gravitational search algorithm from two aspects:improvement and application.Firstly, the basic principle and the implementation procedure of gravitational searchalgorithm are described in detail. The current research status at home and abroad ofgravitational search algorithm is summarized.Secondly, aiming to overcome the premature convergence issue of gravitational searchalgorithm, an improved gravitational search algorithm with complex method is developed. Inthe early period of the search process, gravitational search algorithm is used to globally searchand the gravitational coefficient of gravitational search algorithm is changed to improve theglobal convergence. In the later stage of the search process, complex method is applied toavoid premature convergence. The strong local search ability of the complex method is usedto help the population jump out of the local optimal solution. Through the case study ofstandard functions, the simulation results show that the improved algorithm is feasible andefficient in nonlinear optimization.The improved gravitational search algorithm is then applied to single-objectiveoptimization problem：parameter optimization of support vector machine (SVM). The targetof this method is to minimize the mean square error of output. The strong search capability ofthe improved gravitational search algorithm is used to quickly search for the optimumparameters of SVM. The proposed method is applied to glutamate concentration forecastingin the glutamic acid fermentation process. The simulation result indicates that the method canimprove the forecast accuracy of glutamate concentration and has short training time.Finally, a new multi-objective optimization algorithm based on gravitational searchalgorithm is proposed to solve the complex multi-objective optimization problem. Thealgorithm stores the non-dominated individuals with the elitism strategy, updates thepopulation with the fast non-dominated sorting strategy and crowding distance comparison operator. The individual quality is calculated based on the rank of population which has beensorted by the fast non-dominated sorting strategy. The gravitational coefficient ofgravitational search algorithm is changed to improve the global convergence. Meanwhile theadaptive mutation operator is used to maintain the diversity of population. Simulation resultsof the classical test functions show that the multi-objective gravitational search algorithm hasstrong optimal performance, the final Pareto front is closer to the real Pareto front and thedistribution of Pareto front is more uniform.The multi-objective gravitational search algorithm is then applied to multi-objectiveoptimization of penicillin fermentation process. In this method, the feed rate is optimizationvariable, the maximum of final product and the minimum of substrate consumption areoptimization goals. The establishment of the mathematical model of multi-objective optimumproblem is based on the simplified dynamic model of penicillin. The multi-objectivegravitational search algorithm is used to solve the established mathematical model. Thismethod is compared with the way of constant feed rate. The test results show that the methodin this paper has a higher ratio of the total penicilin production and the total substrateconsumption, indicate the effectiveness of this method in the multi-objective optimization offermentation process.