Multiquadric方法在中尺度气象资料客观分析中的应用

利用浙江省中尺度自动站的温度资料,引入已广泛应用于测绘、地球物理、地图绘制等领域的Multiquadric插值法,并与目前气象领域内采用较多的Cressman方法进行比较,分析两者在不同站点分布和数量下对实况场的逼近程度,并对Multiquadric中的自由设定参数进行讨论。结果表明,总体而言,Multiquadric方法的均方根误差小于Cressman方法。当站点数为100时,两者差异达0.3℃;当站点数1000时,Multiquadric方法相对Cressman的改进已经1%,两者差异非常小。在站点数较少或是站点间隔较大的情况下,Cressman方法较易产生虚假中心,而Multiquadric方法则更加接近实况。Multiquadric平滑参数λ可以被认为是一种低通滤波器,它将高频波滤去,留下较大尺度的系统,同时冷暖空气温度中心的强度随着平滑强度的增加也逐渐减弱。Multiquadric参数c存在一个上限值,当c大于或小于该值时,客观分析结果表现出不同的敏感度。     

Properties of plasma radiation during discharges with improved confinement on HL-2A Tokamak

In the recent experiment on the HL-2A tokamak, two types of improved confinement regimes have been achieved in different configurations. One is the improved confinement regime in limiter configuration during electron cyclotron resonant heating (ECRH), characterized by a sharp decrease in H_α emission accompanied by an increase in the total radiation of plasma, the line averaged electron density and the stored energy of plasma. The other is high confinement regime (H-mode) in divertor configuration during a combination of ECRH and Neutral beam injection (NBI) heating, characterized with edge localized modes (ELMs) besides the features mentioned above. The ELMs are found to be localized on the plasma edge (r/a ≥ 0.8), causing average losses of particles and stored energy in the ranges of about 1--3% and 3--5% respectively during a single ELM event. So far, the ELMs observed in the HL-2A are type III ELMs with low amplitude and high repetition frequency in a range from 200 Hz to 350 Hz. An investigation of the radiated power density profiles shows that radiative cooling effect plays a significant role in the transition back to the L-mode and the triggering of ELM events.     

基于深度学习的地层沉积正演模拟代理模型构建与应用

地层沉积正演模拟方法能更真实地刻画地下地质体的分布规律,比传统的基于地质统计学的建模更有优势,但是条件化难度大,使其应用在实际油藏地质建模时面临较大挑战。地层沉积反演模拟提高了该方法的实用性,基于模拟结果与观测数据定量比较,地层沉积反演模拟在全局优化算法的驱动下不断修正地层沉积正演模拟输入参数,使模拟结果与观测数据吻合度达到最佳。由于反演系统优化参数多,非线性极强,收敛难度大,需要的迭代次数多,单次正演模拟耗时长,导致该方法效率较低。地层沉积反演模拟和深度学习算法中的生成对抗网络相结合,提出了构建地层沉积正演模拟代理模型的方法。以大量的碳酸盐岩地层沉积模拟的合成模型作为样本,通过神经网络训练,形成地层正演模拟器的代理模型,然后将其代入地层沉积反演模拟系统,避免了直接地层正演模拟的长耗时,加快了反演模拟的收敛速度。通过巴哈马西缘碳酸盐岩地层沉积模拟验证了该方法的可行性,采用学习后的生成网络进行沉积反演模拟,反演效率得到了大幅提升。尽管本文展示的是二维实例,也有望扩展应用在三维模型上。