基于模糊贝叶斯网络算法的智能轮椅避障

Obstacle Avoidance System for Intelligent Wheelchair Based on Fuzzy Bayesian Neural Network

摘要:

对传统BP神经网络模糊逻辑的智能轮椅避障方法在训练过程中存在的过拟合和避障路径不够优化的问题,提出了一种模糊贝叶斯网络避障算法以降低神经网络的复杂度;该算法利用模糊神经网络对隶属度函数的参数进行自主学习调整,同时为增强神经网络的泛化能力和计算能力,在网络目标函数中加入权衰减项,利用贝叶斯原理优化神经网络的结构和权值;仿真和实机实验表明,该算法在训练结果和避障效果上均优于传统BP神经网络,提高了智能轮椅避障的实时性,优化了避障路径,可满足用户对智能轮椅安全性和舒适性的需求。 

Abstract:

To solve the over-fitting problem caused by traditional obstacle avoidance method of intelligent wheelchair based on fuzzy logic during training process and the obstacle avoidance path is not optimized, we propose a new obstacle avoidance algorithm to reduce the complexity of the neural network in the training process with fuzzy Bayesian network. Fuzzy neural network is utilized to adjust parameters of membership functions. In order to obtain the ability of good generalization and accurate computing, a penalty term is introduced to the objective function to optimize the structure and the weights of neural networks using Bayesian method. Simulation and physical experiments show that this algorithm is better than the traditional BP network in the training process and the obstacle avoidance path is optimized to meet the users’ needs of the comfort and security better.