基于蚁群BP神经网络的汽车排气噪声有源控制

Active Noise Control of Engine Exhaust Based on AMMAS-BP Algorithm

BP算法容易陷入局部极小值,对初值设置敏感,以及学习速度慢等缺陷,而蚁群算法具有全局寻优、正反馈以及分布式计算等特点,提出一种蚁群BP神经网络混合训练方法(AMMAS-BP)。采用自适应最大-最小蚁群算法(AMMAS),对BP网络的权值参数进行全局训练,再使用BP算法对其进行局部学习。建立基于AMMAS-BP算法的汽车排气噪声有源控制系统的仿真模型。仿真结果表明,该方法改善了BP算法的收敛速度和收敛精度,提高了控制系统的降噪效果。

In this paper, an improved BP algorithm based on adaptive max -min ant system, AMMAS -BP,is presented, which aiming at defecting that BP algorithm is easy to fall into local minimum, sensitive to the initial value and slows to learn, and advantages that ant colony optimization has characteristics of global optimization, positive feedback and distributed computer system. AMMAS is applied to global training of parameters of the neural network before LM algorithm is used for local learning. And the simulation of adaptive and active noise control system to cancel engine exhaust noise is built. The result shows that the method improves convergence speed, accuracy of BP algorithm and noise cancellation effect of the control system.