Improved Gravitational Search Algorithm And Applied In Flatness Pattern Recognition

Gravitational Search Algorithm (GSA) is a global heuristic algorithm which is proposed by Rashedi in2009. Although it has been proved that GSA performs better than Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) in most instances, GSA always has its own defects. So it is necessary to improved the Gravitational Search Algorithm. Strip steel is widly used in national production. With the development of the manufactrue, the flatness of the strip steel is required more and more strict quality. Flatness pattern recognition is a key precondition in flatness control system, because its results directly affect the quality of flatness control. The main research results of this paper can be summarized as follows:Firstly, To overcome the defects and improve the exploration and exploitation abilities, we employ two operators (operator a and b) in the standard Gravitational Search Algorithm. Operator a simulates the evolution rule of population in nature to improve the evolution speed of the population; operator b keeps the accuracy of the population not descent when improving the exploitation of the algorithm. Two operators are combined to improve the ability of GSA. We used23benchmark functions to test the improvement GSA (IGSA). The obtained results confirmed the high performance of the proposed method in solving these various nonlinear functions.Secondly, In the light of the problems existed in flatness pattern recognition models which had low generation ability, slow training speed and poor anti-interference ability, this dissertation proposed a new support vector regression (SVR) model, which is based on linear, quadratic, cubic and biquadrate Legendre orthogonal polynomials. To improve the accuracy of the recognition, the IGSA is import into the SVR model, which is called IGSA-SVR model. Experimental results show that the proposed model not only has fast training speed and high accuracy results, but also has strong generation and anti-interference abilities. The recognition results can provide reliable basis for the formulation of flatness control strategy.