改进粒子BP神经网络在变电站噪声控制中的应用

为了改善变电站噪声控制中已有自适应降噪滤波算法的自适应能力差、收敛速度慢等弊端,提出了一种新的基于粒子群优化（PSO）的误差反向传播神经网络（BPNN）智能滤波算法。该算法针对PSO算法易出现无法兼顾局部、全局搜索和群体多样性丢失等问题,采用以粒子“亲密”度为依据来自适应调整粒子惯性因子和变异率的改进策略;利用该改进粒子群优化（IPSO）算法取代梯度下降算法,实时优化BPNN的权、阈值,使噪声迅速降低,再用梯度下降算法对BPNN的权、阈值作进一步的精细优化,使噪声得到更大程度上的抑制。文中以某变电站变压器噪声信号为仿真声源,分别利用所提算法、PSO-BPNN算法及BPNN算法对该声源信号进行主动抑制,结果表明所提算法性能明显优于另外2种算法的性能,使变压器降噪系统性能得到较大的改善。

The Application of Improved Particle BP Neural Network for Substation Noise Control

In order to overcome the shortcomings of existing adaptive noise filtering algorithms for substation noise control, such as poor adaptive ability and slow convergence speed etc., a new intelligent filter algorithm is proposed based on the error back propagation neural network (BPNN) of particle swarm optimization (PSO). In view of the existing problems of PSO, including difficult reconciliation of local and global search, and the population diversity loss, an improved strategy is used, which uses the particle’s“close” degree to adaptively adjust particle inertia factor and mutat ion rate. The improved particle swarm optimization algorithm is used to replace the gradient descent algorithm for real-time optimization of the weights and thresholds of BPNN, and rapid noise reduction. The gradient descent algorithm is then used for further optimization of the weights and thresholds of BPNN, and as a result, the noise is further subdued. By using the noise signal of a substation transformer for simulation of the sound source, the proposed algorithm, PSO-BPNN algorithm and BPNN algorithm are respectively applied to suppress the sound source. The results show that the performance of the proposed algorithm is superior to the performance of the other two algorithms, and the transformer noise reduction system is improved notably.