基于改进粒子群算法的柔性神经树优化①

神经树采用树结构编码,具有非常好的预测能力和函数逼近能力。模型中的相关参数通常用粒子群优化算法来优化,可是传统的粒子群算法具有容易陷入局部最优值,并且进化后期的收敛速度慢、精度低等缺点,因此会影响神经树的性能。将一种新的改进的粒子群优化算法应用到神经树模型中,并与传统的粒子群算法在柔性神经树的应用比较,表明该改进粒子群算法具有更好的收敛精度,从而改善了神经树的性能。

Optimization of Flexible Neural Tree Based on Improved Particle Swarm

The Neural Tree uses a tree structure coding. It has good predictive ability and function approximation capabilities. In the model, parameters are usually optimized with particle swarm optimization algorithm, but the traditional particle swarm algorithm has following shortcomings like being easily trapped in local optimal value, being slow and having low accuracy in convergence in the later period of the evolution. It affects the performance of neural tree. This paper applies a new improved particle swarm optimization algorithm to the neural tree model, and compares it with the traditional particle swarm algorithm in the application of flexible neural tree. It shows that the improved particle swarm algorithm has better convergence accuracy, thus to improve the performance of the flexible neural tree.