摘要
为了提高Adaline神经网络谐波分析方法对频率波动信号的分析精度,提出了增强型Adaline神经网络模型.该算法将基波频率作为待定的权值,可以同时估计信号频率及各次谐波的幅值和相位,在学习算法中采用动量项方法和频率延迟调整策略以提高算法的收敛性能.讨论了学习率和动量因子对算法收敛性的影响,并给出了各参数的优化设置方法.Matlab仿真结果表明,增强型Adaline谐波分析算法不会产生频谱泄漏,具有较高的分析精度和较快的收敛速度.增强型Adaline谐波分析算法适合于短数据非同步采样下的谐波分析.
An enhanced Adaline neural network model was proposed to improve the accuracy of Adaline neural network harmonic analysis approach for frequency fluctuating signals. The fundamental frequency was treated as weight to be adjusted to estimate the signal frequency and all harmonics＇ amplitudes and phases. The momentum and the delayed frequency adjustment were adopted in the learning algorithm to improve the convergence performance. The influence of learning rates and momentum to the algorithm performance was detailed and the optimum setting of the parameters was revealed. The Matlab simulation results demonstrated that the algorithm achieved high accuracy and rapid convergence with no spectral leakage. The enhanced Adaline harmonic analysis approach is suitable for the situation of short asynchronous sampling data.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2009年第1期166-171,共6页
Journal of Zhejiang University：Engineering Science
