We statistically validate the 2011-2022 earthquake prediction records of Ada, the sixth finalist of the 2nd China AETA in 2021, who made 147 earthquake predictions (including 60% of magnitude 5.5 earthquakes) with a p...We statistically validate the 2011-2022 earthquake prediction records of Ada, the sixth finalist of the 2nd China AETA in 2021, who made 147 earthquake predictions (including 60% of magnitude 5.5 earthquakes) with a prediction accuracy higher than 70% and a confidence level of 95% over a 12-year period. Since the reliable earthquake precursor signals described by Ada and the characteristics of Alfvén waves match quite well, this paper proposes a hypothesis on how earthquakes are triggered based on the Alfvén (Q G) torsional wave model of Gillette et al. When the plume of the upper mantle column intrudes into the magma and lithosphere of the soft flow layer during the exchange of hot and cold molten material masses deep inside the Earth’s interior during ascent and descent, it is possible to form body and surface plasma sheets under certain conditions to form Alfven nonlinear isolated waves, and Alfven waves often perturb the geomagnetic field, releasing huge heat and kinetic energy thus triggering earthquakes. To explain the complex phenomenon of how Ada senses Alvfen waves and how to locate epicenters, we venture to speculate that special magnetosensory cells in a few human bodies can sense earthquake precursors and attempt to hypothesize an algorithm that analyzes how the human biological nervous system encodes and decodes earthquake precursors and explains how human magnetosensory cells can solve complex problems such as predicting earthquake magnitude and locating epicenters.展开更多
液体状态机(Liquid State Machine,LSM)具有实时计算和仿生的特点,在处理时间序列数据上具有巨大潜力。为了研究如何提高神经网络模型训练性能,降低计算复杂度,文章首先梳理和回顾了近几年相关研究文献,其次提出硬件实现和软件模型两个...液体状态机(Liquid State Machine,LSM)具有实时计算和仿生的特点,在处理时间序列数据上具有巨大潜力。为了研究如何提高神经网络模型训练性能,降低计算复杂度,文章首先梳理和回顾了近几年相关研究文献,其次提出硬件实现和软件模型两个优化思路,并总结了不同优化方法的优势与不足,硬件和软件上的优化可以提高神经网络模型学习性能和训练速度,但依然存在可控性差、算法最优解未知等问题,最后针对以上问题对未来的研究方向进行了展望,可为时间序列数据处理和模式识别领域提供优化思路。展开更多
以河南省商丘市为研究区,首先采用OAT(One-at-a-time)方法对WheatGrow模型的输入品种参数进行敏感性分析,在此基础上以抽穗期的开始日期作为约束条件构建代价函数,引入SCE-UA(Shuffled complex evolution method developed at the Unive...以河南省商丘市为研究区,首先采用OAT(One-at-a-time)方法对WheatGrow模型的输入品种参数进行敏感性分析,在此基础上以抽穗期的开始日期作为约束条件构建代价函数,引入SCE-UA(Shuffled complex evolution method developed at the University of Arizona)算法求解得到最优作物品种参数组合,并利用2015—2016年度和2016—2017年度田间实验资料对SCE-UA算法的有效性进行验证。结果表明,基本早熟性参数对穗分化期的模拟结果影响最显著,温度敏感性参数比光周期敏感性参数和生理春化时间参数具有更高的敏感度,生理春化时间的敏感度最低。基于优化后的参数得到的穗分化期模拟值与观测值之间的平均绝对误差(Mean absolute error,MAE)和均方根误差(Root mean square error,RMSE)均小于3 d,表明SCE-UA算法可以有效地获取WheatGrow模型最优品种参数组合。本研究可为WheatGrow模型品种参数的调整优化和模型的推广应用提供依据。展开更多
文摘We statistically validate the 2011-2022 earthquake prediction records of Ada, the sixth finalist of the 2nd China AETA in 2021, who made 147 earthquake predictions (including 60% of magnitude 5.5 earthquakes) with a prediction accuracy higher than 70% and a confidence level of 95% over a 12-year period. Since the reliable earthquake precursor signals described by Ada and the characteristics of Alfvén waves match quite well, this paper proposes a hypothesis on how earthquakes are triggered based on the Alfvén (Q G) torsional wave model of Gillette et al. When the plume of the upper mantle column intrudes into the magma and lithosphere of the soft flow layer during the exchange of hot and cold molten material masses deep inside the Earth’s interior during ascent and descent, it is possible to form body and surface plasma sheets under certain conditions to form Alfven nonlinear isolated waves, and Alfven waves often perturb the geomagnetic field, releasing huge heat and kinetic energy thus triggering earthquakes. To explain the complex phenomenon of how Ada senses Alvfen waves and how to locate epicenters, we venture to speculate that special magnetosensory cells in a few human bodies can sense earthquake precursors and attempt to hypothesize an algorithm that analyzes how the human biological nervous system encodes and decodes earthquake precursors and explains how human magnetosensory cells can solve complex problems such as predicting earthquake magnitude and locating epicenters.
文摘液体状态机(Liquid State Machine,LSM)具有实时计算和仿生的特点,在处理时间序列数据上具有巨大潜力。为了研究如何提高神经网络模型训练性能,降低计算复杂度,文章首先梳理和回顾了近几年相关研究文献,其次提出硬件实现和软件模型两个优化思路,并总结了不同优化方法的优势与不足,硬件和软件上的优化可以提高神经网络模型学习性能和训练速度,但依然存在可控性差、算法最优解未知等问题,最后针对以上问题对未来的研究方向进行了展望,可为时间序列数据处理和模式识别领域提供优化思路。
文摘以河南省商丘市为研究区,首先采用OAT(One-at-a-time)方法对WheatGrow模型的输入品种参数进行敏感性分析,在此基础上以抽穗期的开始日期作为约束条件构建代价函数,引入SCE-UA(Shuffled complex evolution method developed at the University of Arizona)算法求解得到最优作物品种参数组合,并利用2015—2016年度和2016—2017年度田间实验资料对SCE-UA算法的有效性进行验证。结果表明,基本早熟性参数对穗分化期的模拟结果影响最显著,温度敏感性参数比光周期敏感性参数和生理春化时间参数具有更高的敏感度,生理春化时间的敏感度最低。基于优化后的参数得到的穗分化期模拟值与观测值之间的平均绝对误差(Mean absolute error,MAE)和均方根误差(Root mean square error,RMSE)均小于3 d,表明SCE-UA算法可以有效地获取WheatGrow模型最优品种参数组合。本研究可为WheatGrow模型品种参数的调整优化和模型的推广应用提供依据。
