采用传统方法对多峰Brillouin散射谱进行拟合的过程中, 通常是以谱线最大功率点为基准的, 却忽略了其他比该点小但却是极值的功率点。 这样获得的拟合曲线通常只有一个峰值, 相当于把除最高峰之外还有多个小峰的多峰Brillouin散射谱进行了简化, 导致大量有用信息的丢失。 为了提高Brillouin散射谱的特征提取精度, 提出了一种基于MCDM和PSO-LM混合优化算法的多峰Brillouin散射谱特征提取方法(MCDM-PSO-LM)。 MCDM可以识别和准确定位多峰Brillouin散射谱的各个波峰和波谷; PSO-LM混合优化算法可以实现分别对各个波峰和波谷的曲线进行拟合并找到每一个波峰的中心频率, 该算法既克服了PSO算法过早收敛于局部极值和LM算法依赖初值的问题, 又可以将PSO算法的全局搜索能力和LM算法的局部收敛能力结合在一起。 较传统算法而言, MCDM-PSO-LM算法保证了对最优值求解的速度和精度, 提高了运算能力, 使解析解最大限度地接近最优值。 分别在不同信噪比和不同线宽条件下进行仿真验证, 频移和温度误差分析结果表明, MCDM-PSO-LM方法可以对多峰Brillouin散射谱的各个波峰与波谷进行准确定位, 可用于多峰Brillouin散射谱的特征提取, 识别效果明显强于传统算法, 提高了信息分析的准确性。

As to fitting the multi-peaks Brillouin scattering spectrum with traditional method, the maximum power point is usually selected as the benchmark while other extreme value points which are less than the maximum power are lost. The fitting curve only has one peak because the multi-peaks Brillouin scattering spectrum is simplified into the highest peak and several small peaks. So it will lead to the loss of useful information. In order to improve the feature extraction accuracy of Brillouin scattering spectrum, a hybrid optimization algorithm named MCDM-PSO-LM algorithm is presented based on MCDM and PSO-LM algorithm. The MCDM algorithm can identify and locate the peaks and valleys of multi-peaks Brillouin scattering spectrum accurately. The PSO-LM hybrid algorithm can realize the curve fitting on every peak and valley, and it can seach the center frequency shift of each peak. The PSO-LM hybrid algorithm can solves these disadvantages, which PSO algorithm premature convergence to local minimum and LM algorithm depends on the initial value problem. It can also combine the global search ability of PSO algorithm and the local search ability of LM algorithm. Compared with traditional algorithms, MCDM-PSO-LM algorithm can ensure the solving speed and accuracy to the optimal value, and the analytical solution will be close to the optimal value sufficiently. So it improves the operation ability. With different signal to noise ratio and linewidth, the results of frequency shift and temperature error show that the MCDM-PSO-LM method can locate every peak and valley of multi-peaks Brillouin scattering spectrum accurately. Thus, it can be used for the feature extraction of multi-peaks Brillouin scattering spectrum. The recognition effect of this method is obviously better than that of traditional algorithms and it can improve the accuracy of information analysis.