Improved Particle Swarm Optimization Algorithm And Its Application In Neural Networks

Particle swarm optimization(PSO)algorithm is a population-based random optimization algorithm,and it has been successively applied to areas such as electromagnetic optimization,power system etc.However,there are still many theoretical problems and practical applications of particle swarm optimization algorithm that need further study.In order to further improve the convergence speed and accuracy of the algorithm,this dissertation improves the standard particle swarm optimization algorithm mainly from two aspects:algorithm's parameters and update formulas.Next,the improved particle swarm optimization algorithm is used to optimize the grey neural network's weights and thresholds,and then the optimized gray neural network model is applied to grain production predicting problem.The main work is briefly described as follows:(1)A simplified mean particle swarm optimization algorithm with dynamic adjustment of inertia weight(DSMPSO)is proposed.Based on the simplified particle swarm optimization algorithm,the individual and global optima are replaced by their linear combination.Furthermore,the inertia weight is constructed based on Cosine function,the Beta distribution is added into the formula of inertia weight,and thereby the inertia weight can be dynamically adjusted to improve the ability of exploitation and exploration,increasing the diversity of the particle swarm.(2)Combining the improvement idea of the inertia weight and the updated formula in the DSMPSO algorithm,an improved particle swarm optimization algorithm based on S-shaped functions is proposed.The algorithm takes advantage of the characteristics of an upside-down S-shaped function to adjust the inertia weight and introduces an S-shaped function in the position updating equation of the PSO algorithm,so as to enhance the global search ability and efficiency of the algorithm.Based on the SIPSO,an S-shaped function based adaptive particle swarm optimization algorithm(SAPSO)is proposed.The ratio of the individual particle's fitness value to the swarm's average fitness value is used to adaptively adjust the step size in the search process and thereby to enhance the efficiency of the algorithm.Experimental results show that the SAPSO algorithm performs better in terms of the convergence rate and the solution accuracy.(3)A grey neural network prediction model based on improved particle swarm optimization algorithm is established and used for the grain production predicting problem.The radial basis function and the Cauchy distribution function are used respectively to improve the inertia weight update formula and the position update formula of the particle swarm optimization algorithm.The learning factors are regulated using nonlinear change strategy to improve the learning ability of the particles.Next we present an improved particle swarm optimization algorithm to optimize the weights and thresholds of the grey neural network,and the optimized gray neural network model is applied to grain production predicting problem.The simulation results show that the proposed method can achieve higher optimization accuracy than several other methods.