Research On Swarm Intelligence Algorithm Based On Social Learning

Swarm Intelligence Algorithm is a kind of bionic optimization algorithm inspired by natural predation and reproduction behavior.At present,the improvement of intelligent algorithm is mainly in changing convergence factor,inertia weight,population initialization and mixing other algorithms.Social learning theory is a psychological theory developed by the American psychologist Albert Bandura,which focuses on the performance of individuals in a group.In essence,it is the probability that the individual will change his behavior under the influence of reinforcement and punishment.This paper analyzes social learning theory and builds mathematical models,takes the theory in applying this theory to whale optimization algorithms.Then apply to 22benchmark functions and 18 offset functions to verify the effectiveness of the improved algorithm and determine the scope of application of the algorithm.First,set the learning probability function,learning factor,convergence factor and ideal learning individual through experiment and theoretical analysis.In the global exploration and local development,complete the selection of the reference whale and the learning behavior from the ideal learning individual.Secondly,the superiority of SLWOA is proved by comparing GA,PSO,WOA and WOAWC with 7 single-peak and 15 multi-peak benchmark functions respectively.Finally,aiming at the best value in origin,six offsets in Sphere,Rosenbrock,Griewank benchmark functions are run separately for 30 times,which proves that SLWOA is more suitable for solving the problems of single-peak variable dimension and multi-peak variable dimension.In this paper,SLWOA is applied to KDD99 data set to classify.First,the hidden layer data center of RBF neural network is selected by K-means clustering,and set parameters of gauss radial function.secondly,optimize the weights of hidden layer and output layer with SLWOA.Then the optimized neural network is applied to KDD99Dataset.Finally,the accuracy,false positive rate and false negative rate of GRD-RBF,WOA-RBF and SLWOA-RBF were compared.The experimental results show that the performance of SLWOA-RBF neural network is better than that of the original RBF neural network on the test data set.