Abstract:

We presented a hybrid learning algorithm for radial basis function neural network, which is based on particle swarm optimization, K-means clustering and subtractive clustering algorithm. The algorithm can be used to determine the number of hidden layer nodes adaptively by using subtractive clustering, avoiding the manmade interference for adjusting the hidden layer
nodes. The initial particle swarm of particle swarm optimization can be formed by K-means clustering algorithm to avoid the randomness of the initial particle swarm. In this way, the optimization capability of particle swarm optimization is improved. Particle swarm optimization algorithm is used to train all the parameters. Numerical results show that the accuracy of the improved hybrid algorithm is superior to two existing popular methods.

Key words: clustering, particle swarm optimization, radial basis function(RBF) neural network