Power spectral analysis is discussed and an exmple of its application to a geochemical stratigraphic profile is presented.In the Dachang area,all major elements,as well as the maximal sedimentary trend factor(MSTF),de...Power spectral analysis is discussed and an exmple of its application to a geochemical stratigraphic profile is presented.In the Dachang area,all major elements,as well as the maximal sedimentary trend factor(MSTF),demonstrate consistent change periodicity,but are out-of-step with cach other.This can be explained by sedimentogecochemical consideration.The results of power spectral analysis enable geochemists to group elements on the basis of the distribution of their changed periods,and to interpret some special geological and geochemical situations.展开更多
On the basis of the analytical results of the period components of monthly mean sea level of 236 stationsin the Pacific, the period components plus linear trend are ed to fit the monthly mean sea level series. The sta...On the basis of the analytical results of the period components of monthly mean sea level of 236 stationsin the Pacific, the period components plus linear trend are ed to fit the monthly mean sea level series. The statisticalresults of linear trend Coefficients of these stations indicate that, if the abnormal values of sea-level rise and fall are neglected, the average rise rate of relative sea level in the Pacific is 1. 16 mm/a. Affected by nonuniformity Of land subsidence and other factors, the regional change of relative sea level rise or fall in the Pacific is greater. In the light of thepositive or negative values of linear trend coefficients as well as the geographical position of the sea area, zoning is madeof the sea level rise or fall in the Pacific including the coastal areas of China and Southeast Asia to obtain the averagerate of rise or fall in each sea area. The rise or fall trends of relative sea level obtained for the entire Pacific Ocean,west coast of North America, the northern and central South America, the greater part of the tropical Pacific and thecoastal Islands of Japan are basically in keeping with the other relevant results. The regional average estimated result ofthe relative sea level in the coast of East Asia is on the rise while the estimated results provided by Barnett tend todrop; the main cause of this nonuniformity is the number of stations selected and the distributional density.展开更多
This paper summarizes the general methods,existing problems and their causes of the period analysis for the monthly mean sea level and points out that it is the key to the analysing period signals and forecasting the ...This paper summarizes the general methods,existing problems and their causes of the period analysis for the monthly mean sea level and points out that it is the key to the analysing period signals and forecasting the change trend of the monthly mean sea level that the periods of the signals are selected reasonably. As there are often many period signals in these series, nonlinear effects exist between pairs of period signals. In order to avoid the false periods that may be introduced due to the effects of side lobes and the periods with statistical phase significance coherence that may be introduced due to the effects of nonlinear effects and their restraint to other period signals, the maximum entropy spectral analysis and the corresponding significance period test may be performed repeatedly on the basis of the bispectrum analysis and meanwhile the most significant period component is filtered out by the least square filtering method, i. e., the method of the significance period analysis with mixed spectra modeled by a nonlinear system is adopted and the signal periods approaching the reality are selected one by one. The examples of the bispectrum analysis, the signal period analysis by mixed spectra and the fitting parameters for combined period components with linear trend in the time series of monthly mean sea level are given in this paper.展开更多
提出基于强化学习三态组合长短时记忆神经网络(reinforcement learning 3-states combined long and short time memory neural network,简称RL-3S-LSTMNN)的旋转机械状态退化趋势预测新方法。笔者提出的RL-3SLSTMNN中,采用最小二乘线...提出基于强化学习三态组合长短时记忆神经网络(reinforcement learning 3-states combined long and short time memory neural network,简称RL-3S-LSTMNN)的旋转机械状态退化趋势预测新方法。笔者提出的RL-3SLSTMNN中,采用最小二乘线性回归方法构造单调趋势识别器,将旋转机械整体的状态退化趋势分为平稳、下降、上升3种单调的趋势单元,并通过强化学习为每一种单调趋势单元选择一种隐层层数和隐层节点数与之相适应的长短时记忆神经网络,提高了RL-3S-LSTMNN的泛化性能和非线性逼近能力,使所提出的状态退化趋势预测方法具有较高的预测精度。用不同隐层数、隐层节点数和3种单调趋势单元分别表示Q表的动作和状态,并将长短时记忆神经网络(long and short time memory neural network,简称LSTMNN)输出误差与Q表的更新相关联,避免了决策函数的盲目搜索。结果表明:提高了RL-3S-LSTMNN的收敛速率,使所提出的预测方法具有较高的计算效率;滚动轴承状态退化趋势预测实例验证了该方法的有效性。展开更多
文摘Power spectral analysis is discussed and an exmple of its application to a geochemical stratigraphic profile is presented.In the Dachang area,all major elements,as well as the maximal sedimentary trend factor(MSTF),demonstrate consistent change periodicity,but are out-of-step with cach other.This can be explained by sedimentogecochemical consideration.The results of power spectral analysis enable geochemists to group elements on the basis of the distribution of their changed periods,and to interpret some special geological and geochemical situations.
文摘On the basis of the analytical results of the period components of monthly mean sea level of 236 stationsin the Pacific, the period components plus linear trend are ed to fit the monthly mean sea level series. The statisticalresults of linear trend Coefficients of these stations indicate that, if the abnormal values of sea-level rise and fall are neglected, the average rise rate of relative sea level in the Pacific is 1. 16 mm/a. Affected by nonuniformity Of land subsidence and other factors, the regional change of relative sea level rise or fall in the Pacific is greater. In the light of thepositive or negative values of linear trend coefficients as well as the geographical position of the sea area, zoning is madeof the sea level rise or fall in the Pacific including the coastal areas of China and Southeast Asia to obtain the averagerate of rise or fall in each sea area. The rise or fall trends of relative sea level obtained for the entire Pacific Ocean,west coast of North America, the northern and central South America, the greater part of the tropical Pacific and thecoastal Islands of Japan are basically in keeping with the other relevant results. The regional average estimated result ofthe relative sea level in the coast of East Asia is on the rise while the estimated results provided by Barnett tend todrop; the main cause of this nonuniformity is the number of stations selected and the distributional density.
文摘This paper summarizes the general methods,existing problems and their causes of the period analysis for the monthly mean sea level and points out that it is the key to the analysing period signals and forecasting the change trend of the monthly mean sea level that the periods of the signals are selected reasonably. As there are often many period signals in these series, nonlinear effects exist between pairs of period signals. In order to avoid the false periods that may be introduced due to the effects of side lobes and the periods with statistical phase significance coherence that may be introduced due to the effects of nonlinear effects and their restraint to other period signals, the maximum entropy spectral analysis and the corresponding significance period test may be performed repeatedly on the basis of the bispectrum analysis and meanwhile the most significant period component is filtered out by the least square filtering method, i. e., the method of the significance period analysis with mixed spectra modeled by a nonlinear system is adopted and the signal periods approaching the reality are selected one by one. The examples of the bispectrum analysis, the signal period analysis by mixed spectra and the fitting parameters for combined period components with linear trend in the time series of monthly mean sea level are given in this paper.
文摘提出基于强化学习三态组合长短时记忆神经网络(reinforcement learning 3-states combined long and short time memory neural network,简称RL-3S-LSTMNN)的旋转机械状态退化趋势预测新方法。笔者提出的RL-3SLSTMNN中,采用最小二乘线性回归方法构造单调趋势识别器,将旋转机械整体的状态退化趋势分为平稳、下降、上升3种单调的趋势单元,并通过强化学习为每一种单调趋势单元选择一种隐层层数和隐层节点数与之相适应的长短时记忆神经网络,提高了RL-3S-LSTMNN的泛化性能和非线性逼近能力,使所提出的状态退化趋势预测方法具有较高的预测精度。用不同隐层数、隐层节点数和3种单调趋势单元分别表示Q表的动作和状态,并将长短时记忆神经网络(long and short time memory neural network,简称LSTMNN)输出误差与Q表的更新相关联,避免了决策函数的盲目搜索。结果表明:提高了RL-3S-LSTMNN的收敛速率,使所提出的预测方法具有较高的计算效率;滚动轴承状态退化趋势预测实例验证了该方法的有效性。
文摘针对变分模态分解(variational mode decomposition,VMD)中模态数K和惩罚因子α无法自适应确定的问题,提出了基于快速变分模态分解(fast VMD,FVMD)的滚动轴承故障特征提取方法。首先,利用频谱趋势分割方法对滚动轴承振动信号进行分析,确定频谱趋势分割边界,进而自适应确定VMD的分解模态数K和惩罚因子α、模态初始中心频率ω;其次,根据参数K、α、ω,完成原始振动信号的自适应分解,并基于有效权重峭度准则提取有效本征模态函数(intrinsic mode function,IMF)分量;最后,利用希尔伯特包络解调计算有效IMF分量重构信号的包络频谱图,完成滚动轴承故障特征的提取。使用仿真信号、美国凯斯西储大学(Case Western Reserve University,CWRU)和美国航空航天局(National Aeronautics and Space Administration,NASA)的滚动轴承数据完成所提方法与传统VMD方法的对比试验。结果表明,所提方法能够自适应确定VMD的分解模态数K和惩罚因子α,提高VMD的计算效率,同时有效提取到滚动轴承的故障特征频率,证明了所提方法的有效性和可行性。
