Propagation of strong femtosecond hyper-Gaussian pulses in a cascade three-level molecular system is studied by solving numerically the Maxwell–Bloch equations by the iterative predictor-corrector finite-difference t...Propagation of strong femtosecond hyper-Gaussian pulses in a cascade three-level molecular system is studied by solving numerically the Maxwell–Bloch equations by the iterative predictor-corrector finite-difference time-domain method.Optical power limiting behavior induced by strong nonlinear two-photon absorption is observed for different orders of the femtosecond hyper-Gaussian pulses. Pulses of a higher order temporal profile are found to have a wider power range of optical limiting but a larger output saturation intensity. Both the output saturation value and the damage threshold of optical power limiting decrease with pulse duration increasing. The decrease of the pulse area along the pulse propagation is much slower than that obtained from the two-photon area theorem due to invalidity of the slowly varying amplitude approximation and the monochromatic field hypothesis.展开更多
超参数优化是减少高斯过程回归(Gaussian process regression,GPR)学习方法计算量,提高高斯模型性能的一个重要问题。为解决超参数优化问题中先验知识匮乏,对初始值过分依赖且易陷入局部最优等问题,文章引入粒子群优化(particle swarm o...超参数优化是减少高斯过程回归(Gaussian process regression,GPR)学习方法计算量,提高高斯模型性能的一个重要问题。为解决超参数优化问题中先验知识匮乏,对初始值过分依赖且易陷入局部最优等问题,文章引入粒子群优化(particle swarm optimization,PSO)算法,并结合差分速度更新公式及自适应变异策略,提出了一种自适应差分粒子群-高斯过程回归优化(adaptive differential particle swarm optimization-Gaussian process regression,ADPSO-GPR)算法对GPR中超参数进行自适应优化。该算法在线性与非线性两类时序回归序列上与多种主流超参数优化算法进行对比,实验结果表明,采用该文算法优化超参数后的GPR具有较高的拟合精度及泛化能力。展开更多
基金Project supported by the National Natural Science Foundation of China(Grant No.11574082)the Fundamental Research Funds for the Central Universities,China(Grant No.2018MS050)
文摘Propagation of strong femtosecond hyper-Gaussian pulses in a cascade three-level molecular system is studied by solving numerically the Maxwell–Bloch equations by the iterative predictor-corrector finite-difference time-domain method.Optical power limiting behavior induced by strong nonlinear two-photon absorption is observed for different orders of the femtosecond hyper-Gaussian pulses. Pulses of a higher order temporal profile are found to have a wider power range of optical limiting but a larger output saturation intensity. Both the output saturation value and the damage threshold of optical power limiting decrease with pulse duration increasing. The decrease of the pulse area along the pulse propagation is much slower than that obtained from the two-photon area theorem due to invalidity of the slowly varying amplitude approximation and the monochromatic field hypothesis.
文摘为解决位置指纹定位算法中指纹采集工作量大、定位精度低的问题,提出一种基于稀疏指纹采集和改进加权K最近邻(weighted k-nearest neighbor,WKNN)的定位算法。稀疏选定参考点并采集来自各接入点(access point,AP)的接收信号强度(received signal strength,RSS),根据容错四分位法对采集的RSS进行异常值预处理;利用经过预处理的指纹数据训练高斯过程回归(Gaussian process regression,GPR)模型,通过共栖生物搜索算法(symbiotic organisms search,SOS)求取模型最优超参数以提高模型的泛化能力,进而预测定位区域内非参考点的RSS;由有限参考点数据通过SOS-GPR模型的训练与预测生成密集位置指纹库,结合由卡方距离和AP加权改进的WKNN算法完成仿真验证。实验结果表明,在保证定位精度的前提下,稀疏指纹采集法较传统全采集法减少50%的采集工作量;与原WKNN算法和M-KWNN算法相比,提出的WKNN算法有效提高了定位精度。
