Suppose {Xn} is a random walk in time-random environment with state space Z^d, ｜Xn｜ approaches infinity, then under some reasonable conditions of stability, the upper bound of the discrete Packing dimension of the r...Suppose {Xn} is a random walk in time-random environment with state space Z^d, ｜Xn｜ approaches infinity, then under some reasonable conditions of stability, the upper bound of the discrete Packing dimension of the range of {Xn} is any stability index α. Moreover, if the environment is stationary, a similar result for the lower bound of the discrete Hausdorff dimension is derived. Thus, the range is a fractal set for almost every environment.展开更多
在实际的自适应滤波系统中普遍存在随机处理延迟和异质测量噪声(如高斯噪声、脉冲噪声等)的问题,而现有的变步长最小均方误差(Variable Step-Size Least Mean Square,VSSLMS)算法在分析时通常假设系统为无延时系统.为了解决上述问题,提...在实际的自适应滤波系统中普遍存在随机处理延迟和异质测量噪声(如高斯噪声、脉冲噪声等)的问题,而现有的变步长最小均方误差(Variable Step-Size Least Mean Square,VSSLMS)算法在分析时通常假设系统为无延时系统.为了解决上述问题,提出一种随机延迟容忍的鲁棒VSSLMS算法,利用Squareplus函数的两个优势:(1)在时延条件下对梯度估计稳定性具有固有平滑性;(2)针对多种类型分布的非线性干扰具有抑制能力.在理论上分析该算法的均方误差和稳态均方误差以评估其性能,并设计系统辨识实验仿真来验证该算法的有效性,且结果与理论分析一致,也优于现有的自适应滤波算法.因此提出的算法不仅表现出更好的稳态性能,在对抗随机时延和多类型测量噪声时也具有更好的鲁棒性.展开更多
使用2021年GRAPES(Global/Regional Assimilation and Prediction System)模式数据和FY-4A卫星数据分析酒泉地区云量时空特征,采用时间自适应方法、动态变参数方法以及随机森林方法建立云量预测模型。结果表明:酒泉及周边地区总云量日...使用2021年GRAPES(Global/Regional Assimilation and Prediction System)模式数据和FY-4A卫星数据分析酒泉地区云量时空特征,采用时间自适应方法、动态变参数方法以及随机森林方法建立云量预测模型。结果表明:酒泉及周边地区总云量日变化幅度不大,季节变化特征明显,春、夏季多,秋、冬季较少,北部云量较少、南部云量多。不同格点的云量受到不同因素的影响,使用动态变参数方法,即根据预报因子和云量相关性在不同格点上动态选取预报因子构建随机森林模型,云量预测准确率为0.55~0.80。采用时间自适应方法使随机森林模型能够更新换代,云量预测准确性在0.55左右,数据量不足导致随机森林模型预测云量的准确率下降。展开更多
In this paper, we give a general model of random walks in time-random environment in any countable space. Moreover, when the environment is independently identically distributed, a recurrence-transience criterion and ...In this paper, we give a general model of random walks in time-random environment in any countable space. Moreover, when the environment is independently identically distributed, a recurrence-transience criterion and the law of large numbers are derived in the nearest-neighbor case on Z^1. At last, under regularity conditions, we prove that the RWIRE {Xn} on Z^1 satisfies a central limit theorem, which is similar to the corresponding results in the case of classical random walks.展开更多
基金Project supported by NNSF of China (10371092)Foundation of Wuhan University
文摘Suppose {Xn} is a random walk in time-random environment with state space Z^d, ｜Xn｜ approaches infinity, then under some reasonable conditions of stability, the upper bound of the discrete Packing dimension of the range of {Xn} is any stability index α. Moreover, if the environment is stationary, a similar result for the lower bound of the discrete Hausdorff dimension is derived. Thus, the range is a fractal set for almost every environment.
文摘在实际的自适应滤波系统中普遍存在随机处理延迟和异质测量噪声(如高斯噪声、脉冲噪声等)的问题,而现有的变步长最小均方误差(Variable Step-Size Least Mean Square,VSSLMS)算法在分析时通常假设系统为无延时系统.为了解决上述问题,提出一种随机延迟容忍的鲁棒VSSLMS算法,利用Squareplus函数的两个优势:(1)在时延条件下对梯度估计稳定性具有固有平滑性;(2)针对多种类型分布的非线性干扰具有抑制能力.在理论上分析该算法的均方误差和稳态均方误差以评估其性能,并设计系统辨识实验仿真来验证该算法的有效性,且结果与理论分析一致,也优于现有的自适应滤波算法.因此提出的算法不仅表现出更好的稳态性能,在对抗随机时延和多类型测量噪声时也具有更好的鲁棒性.
文摘使用2021年GRAPES(Global/Regional Assimilation and Prediction System)模式数据和FY-4A卫星数据分析酒泉地区云量时空特征,采用时间自适应方法、动态变参数方法以及随机森林方法建立云量预测模型。结果表明:酒泉及周边地区总云量日变化幅度不大,季节变化特征明显,春、夏季多,秋、冬季较少,北部云量较少、南部云量多。不同格点的云量受到不同因素的影响,使用动态变参数方法,即根据预报因子和云量相关性在不同格点上动态选取预报因子构建随机森林模型,云量预测准确率为0.55~0.80。采用时间自适应方法使随机森林模型能够更新换代,云量预测准确性在0.55左右,数据量不足导致随机森林模型预测云量的准确率下降。
基金the Natural Science Foundation of Anhui Province （No. KJ2007B122） the Youth Teachers Aid Item of Anhui Province （No. 2007jq1117）.
文摘In this paper, we give a general model of random walks in time-random environment in any countable space. Moreover, when the environment is independently identically distributed, a recurrence-transience criterion and the law of large numbers are derived in the nearest-neighbor case on Z^1. At last, under regularity conditions, we prove that the RWIRE {Xn} on Z^1 satisfies a central limit theorem, which is similar to the corresponding results in the case of classical random walks.
