基于混沌优化的多层前馈神经网络

BP算法是应用广泛的一种多层前馈神经网络模型,针对算法求解精度低、搜索速度慢、易于陷入局部极值点等问题,根据混沌理论的全局优化思想,提出采用"多次载波"技术将混沌优化和前馈神经网络相结合,利用已找到的近似最优解来启发搜索全局最优解的方法训练神经网络,以布尔函数识别、曲线逼近、模式识别3个典型应用对算法进行验证。研究结果表明,算法具有较好的泛化能力和快速全局收敛的性能,特别是针对中小规模的网络,混沌优化算法在训练时间、全局收敛率等指标方面优于BP算法。

Multi-layer feed-forward neural network based on chaos optimization algorithm

BP algorithm is widely used as a multi-layer feed-forward neural network model.Aiming at low precision,long search time,prone to local optimum problems,and according to chaos theory of global optimization idea,chaos optimization and combination of feed-forward neural network based on ＂multiple-carrier＂ technology was proposed.Using the approximate optimal result has been found in heuristic search method of global optimization to train the neural network,boolean function recognition,curve approximation,pattern recognition were validated.The results indicate that the algorithm has better generalization ability and fast global convergence of performance,especially for small and medium-scale networks,chaos optimization algorithm in training time,the global convergence rate are better than BP algorithm.