Study And Application Of An Improved Swarm Optimization Algorithms Based On Social Force Model

Optimization is a kind of applied science which searches the optimal solution of problems in certain conditions. The purpose of optimization algorithm is to solve all kinds of optimization problems in reality. Optimization algorithm can be divided into deterministic algorithm and probabilistic algorithm by its processing methods. The traditional optimization algorithms usually look for objective function’s extremum according to mathematical analysis method. But when dealing with discontinuous and non-differentiable functions, the definitive algorithm is helpless and its global search ability is poor, which restricts the application of definitive algorithm to a great extent. As an important branch of probabilistic algorithm, evolutionary algorithm has become a research hotspot in the field of optimization algorithm. Many researchers have found that different species can solve all kinds of problems by their instinctive social behavior model and have the ability of self-organization, self-learning and adaptive since the establishment of bionics. Inspired by nature, researchers have designed simple, generic swarm intelligence optimization algorithm by simulating different species’ social behaviors.In nature, the populations are made up of multiple agents. The whole population "emergent" complex behavior characteristics by means of every agent’s following the behavior rules of their population and collaboration of multiple agents. And this kind of behavior characteristics of population is not a simple superposition of individual ability. The researchers worked out the swarm intelligence optimization algorithm to solve complex optimization problem through the simulation of different populations’ intelligent behavior. In recent years, the researchers have put forward a lot of typical swarm intelligence optimization algorithms (particle swarm optimization algorithm, artificial fish algorithm, swarm algorithm, ant colony algorithm, etc.) according to the abstract of biological behaviors. Although these intelligent algorithms can obtain satisfactory results in function optimization, they still show different faults in the process of optimization, such as poor search ability, slow convergence speed, premature convergence and search stagnation. Up to now, there is no good algorithm to solve all optimization problems. Therefore, it’s necessary to improve existing algorithms or design optimization algorithm based on different mechanism to solve various optimization problems.Social force model is a simulation model that simulates the pedestrian flow in crowded places. And it has been widely applied in the area of crowd evacuation simulation and analysis, performance evaluation of the construction safety, traffic hub flow analysis. Social force model defines the three forces that pedestrians are suffered:(1) Self-driving force. It shows target’s internal expectation to pedestrian.(2) Force between the pedestrians. It’s showed to avoid congestion when the distance between the pedestrian is very small.(3) The force between pedestrians and buildings. It simulates pedestrain’s psychology to avoid collision with buildings. As the framework of multiple individuals self-driving system, social force model not only describes individuals’thinking and responding ability to the surroundings, but also well abstracts individuals’psychological desire and force condition. It realistically depicts the whole process of individual movement.The existing swarm optimization algorithm based on social force model gains good effect to the optimization problem of low-dimension and multi-objective function. But it still exists many problems in practice, such as poor solving accuracy, slow convergence speed and easily falling into local optimum for high dimensional functions. Some improvement strategies are adopted in this paper to solve some problems of SFSO.63benchmark functions are selected to test the effectiveness of the algorithm. The test results demonstrate that the improved SFSO can achieve high balance in global search and local search, has strong global search ability and has higher precision and higher success rate to each kind of functions.Now, there is no good theoretical basis for the parameters selection of support vector machines (SVM). A large number of experiments demonstrate that the improper parameters selection affects the performance of SVM to a great extent. Thus, we introduce an improved SFSO to the parameter optimization of SVM. The classification performance of the optimized SVM is obviously improved.Nowadays, Deep Belief Networks (DBN) is a popular machine learning method. But when using the DBN to conduct the characteristic learning of the original data, a problem lying in it is the determination of DBN structure parameters. Among it, the selection of the number of hidden layer nodes is a personal determine process, which needs to rely on certain experience. Some research results demonstrate that for particular data sets, DBN is hard to modeling correctly with improper parameter setting. Selecting parameter by experience is not only time-consuming, but also difficult to achieve satisfactory results. Therefore, we use the improved SFSO algorithm to optimize the selection of the number of hidden layer nodes to achieve a better structure of DBN.