Improvement And Convergence Analysis Of Bat Algorithm

Bat algorithm is a new heuristic optimization algorithm simulating the behavior of the bat echolocation.Because of its simple structure and definite practical background,it is widely followed,and has been applied in many fields successfully,such as engineering optimization,data mining.Like many bionics random optimization algorithm,the theoretical foundation of BA is rather weak.It is easy to fall into local optimum,with slowly later convergence speed and other shortcomings.By improving the theory support and performance of BA to extend the algorithm application field,this thesis carried out the following work:1.The bionic principle,mathematical model and basic flow of the bat algorithm are described in detail,and the bat algorithm and other bionic algorithms are analyzed and compared.On this basis,the improvement strategies of this kind of algorithm is summarized.2.The bat algorithm's global convergence is analyzed.By establishing the probability measure space of BA,verifies that this algorithm is a form of contraction mapping,and its orbit is bounded in probability.The global convergence of BA is proved by the fixed point theorem of PM space.3.The bat algorithm is optimized with chaotic disturbance and elite opposition learning.In order to overcome the slow convergence speed and premature phenomenon of BA,the chaotic disturbance strategy is used to resist the sudden decrease of the population diversity and avoid the algorithm falling into a local optimum.At the same time,the chaos and elite opposition learning strategy joined with increases the slow convergence speed.That balances the development ability and detection ability,and improves the performance of the BA.Numerical simulation results show that compared with basic BA,the improved algorithm has higher search accuracy and faster convergence speed.4.The bat algorithm is optimized with a weighting strategy.First,by designing a weighting function based on the ability of optimizing and getting rid of local extremum,bats no longer learn only from the globally optimal bat,but share and exchange information with all bats in the neighborhood.This algorithm is more in line with the reality of biological society and at the same time reduces the risk of the algorithm falling into the local optimum.Second,the frequency is no longer a random number,but adjusted according to the bat's self optimization ability to get or keep the excellent ability.Numerical simulation results show that the new algorithm has higher search accuracy and faster convergence than basic bat algorithm and a bat algorithm based on time-varying inertia weight.5.The questions should be further discussed of bat algorithm and the improved algorithm in this thesis are put forward.