Improved Fruit Fly Optimization Algorithm With Changing Step Based On Random Mechanism And Its Application

With the continuous development of computer technology and the emergence of artificial intelligence,people tend to be more and more automated and intelligent when handling transactions,so many intelligent meta-heuristic optimization algorithms have been generated,Fruit Fly Optimization Algorithm(FOA)is one of them.This Algorithm has the advantages of less control parameters,simple structure,short running time and easy implementation,and has been widely used.Like other meta-heuristic intelligent Optimization Algorithm,FOA is easy to fall into local optimal in the iterative process,in solving high dimensional multiple extremum problems is easy to appear premature convergence phenomenon.In order to compensate for the shortcomings of FOA,this paper proposes improved fruit fly optimization algorithm with changing step based on random mechanism(CSFOA),and uses it to optimize Generalized Regression Neural Network(GRNN)to predict the depth of bullet scratch.The main research contents are as follows(1)This paper proposes improved fruit fly optimization algorithm with changing step based on random mechanism.Adopts the strategy of dynamic decreasing step-size to make the entire population search in a larger spatial range.At the beginning of algorithm iteration,the algorithm has a relatively large search step-size,the algorithm can search in a global range and to some extent enhance the detection ability of the algorithm.As the iteration progresses,the search step-size gradually decreases,so the algorithm can conduct depth search near the current optimal solution to improve the algorithm's optimization accuracy in the later iteration.The dynamic decreasing step-size strategy can compensate for the imbalance between global search and depth detection in the traditional FOA.Once the traditional FOA finds the optimal fruit fly in the visual search process,the entire fruit fly group will fly to the location,which is not necessarily the location of the optimal solution,but probably the local optimal,so the diversity of the entire fruit fly population will decline,unable to conduct the global search and falling into the local optimal.In order to escape from local optimization,random perturbation mechanism is introduced in this paper,which enables the algorithm to jump out of local optimization with a greater probability,it will increase the diversity of solutions,and thus realize global search.(2)Several classical test functions were selected to verify CSFOA.The performance of the algorithm was tested from two aspects.First,the fixed population size and iteration times were used to compare the convergence speed and optimization accuracy of the algorithm.Second,under the fixed convergence precision,the average iteration times and success rate of the algorithm are compared.Simulation results show that CSFOA has higher convergence efficiency,better optimization performance and robustness.(3)The improved FOA optimize the smoothing factor of the GRNN,and use it to test the nonlinear function approximation error.The simulation experiments show that in the approximation of function test,the improved GRNN reduces the function error,improved the precision.(4)Characteristics of bullet scratchs depth prediction can provide reliable basis for investigators to identify a gun,this paper proposes a modeling method of bullet scratchs depth with optimized GRNN based on improved fruit fly algorithm.Compared with GRNN,optimized GRNN based on the traditional FOA and BP neural network,experimental result show that the proposed method has better stability and smaller prediction error.