摘要：The paper proposes an approach to optimize the connection weights and network structure of BP neural network (BPNN) which based on improved particle swarm optimization (PSO) algorithm. For each network structure, the algorithm generates a series of particles which consist of connection weights and threshold values, and selects the best network structure according to the improved PSO algorithm. Because the PSO algorithm is easy to fall into local optimums, the algorithm introduces crossover operator and mutation operator to heighten the ability of jumping the local optimums. Compared with the basic BP algorithm, the results show that performances of the improved PSO-BP algorithm are superior to it, and the paper applies this BPNN model to metallogenic prediction and give the detailed steps.
