大规模新能源并网后电力系统的电压安全稳定问题突出,亟需一种兼具准确性和实用性的方法来评估系统的电压支撑强度。为此,该文提出一种基于贝叶斯深度学习的新能源多场站短路比(multiple renewable energy station short circuit ratio,...大规模新能源并网后电力系统的电压安全稳定问题突出,亟需一种兼具准确性和实用性的方法来评估系统的电压支撑强度。为此,该文提出一种基于贝叶斯深度学习的新能源多场站短路比(multiple renewable energy station short circuit ratio,MRSCR)智能增强方法。首先,聚焦于MRSCR缺乏准确的临界短路比(critical short circuit ratio,CSCR)问题,提出CSCR样本集的构建流程,并据此开发样本的批量仿真程序。然后,利用多门控混合专家网络对各新能源接入点的CSCR进行同步预测,并结合贝叶斯深度学习提升预测精度,量化预测不确定性。最后,考虑到点估计的弊端,提出一种基于动态阈值的不等式方法来给出兼具可靠性和清晰性的区间估计,为不同的决策需求提供多种属性的预测值。在CEPRI-FS-102节点系统上的测试结果表明,所提方法可有效提高电压支撑强度的评估精度和速度,其预测信息可为决策过程提供重要的指导意义。展开更多
采用贝叶斯深度神经网络对液滴模型进行优化改进,并运用KL(Kullback-Leibler)散度与变分推断的方法使得模型便于实现.以最新的原子核数据(AME2020)中2457个有精确值的原子核(Z≥8和N≥8)作为总数据集,随机选取其中80%的数据为训练集用...采用贝叶斯深度神经网络对液滴模型进行优化改进,并运用KL(Kullback-Leibler)散度与变分推断的方法使得模型便于实现.以最新的原子核数据(AME2020)中2457个有精确值的原子核(Z≥8和N≥8)作为总数据集,随机选取其中80%的数据为训练集用于模型训练,通过预测余下的20%进行模型验证.最终两个数据集的误差均方根(RMS)基本相等,而且全部数据的RMS从2.9894 Me V降到0.5695 Me V,下降了80%,呈现出较好的结果.此模型进行了输入参数上的改进(区域限定策略),使得未知核(Z=118—126)可以被限定在一个固定的区域内,从而提高了预测的准确性.为了验证这一性质,对实验数据(Z=100—117)进行了预测计算,结果也与实验值符合得很好.最后使用该方案对未知元素Z=118—126进行了预测,为以后寻找新元素提供了新思路.展开更多
A model of continuous-time insider trading in which a risk-neutral in-sider possesses two imperfect correlated signals of a risky asset is studied.By conditional expectation theory and filtering theory,we first establ...A model of continuous-time insider trading in which a risk-neutral in-sider possesses two imperfect correlated signals of a risky asset is studied.By conditional expectation theory and filtering theory,we first establish three lemmas:normal corre-lation,equivalent pricing and equivalent profit,which can guarantee to turn our model into a model with insider knowing full information.Then we investigate the impact of the two correlated signals on the market equilibrium consisting of optimal insider trading strategy and semi-strong pricing rule.It shows that in the equilibrium,(1)the market depth is constant over time;(2)if the two noisy signals are not linerly correlated,then all private information of the insider is incorporated into prices in the end while the whole information on the asset value can not incorporated into prices in the end;(3)if the two noisy signals are linear correlated such that the insider can infer the whole information of the asset value,then our model turns into a model with insider knowing full information;(4)if the two noisy signals are the same then the total ex ant profit of the insider is increasing with the noise decreasing,while down to O as the noise going up to infinity;(5)if the two noisy signals are not linear correlated then with one noisy signal fixed,the total ex ante profit of the insider is single-peaked with a unique minimum with respect to the other noisy signal value,and furthermore as the noisy value going to O it gets its maximum,the profit in the case that the real value is observed.展开更多
文摘大规模新能源并网后电力系统的电压安全稳定问题突出,亟需一种兼具准确性和实用性的方法来评估系统的电压支撑强度。为此,该文提出一种基于贝叶斯深度学习的新能源多场站短路比(multiple renewable energy station short circuit ratio,MRSCR)智能增强方法。首先,聚焦于MRSCR缺乏准确的临界短路比(critical short circuit ratio,CSCR)问题,提出CSCR样本集的构建流程,并据此开发样本的批量仿真程序。然后,利用多门控混合专家网络对各新能源接入点的CSCR进行同步预测,并结合贝叶斯深度学习提升预测精度,量化预测不确定性。最后,考虑到点估计的弊端,提出一种基于动态阈值的不等式方法来给出兼具可靠性和清晰性的区间估计,为不同的决策需求提供多种属性的预测值。在CEPRI-FS-102节点系统上的测试结果表明,所提方法可有效提高电压支撑强度的评估精度和速度,其预测信息可为决策过程提供重要的指导意义。
文摘采用贝叶斯深度神经网络对液滴模型进行优化改进,并运用KL(Kullback-Leibler)散度与变分推断的方法使得模型便于实现.以最新的原子核数据(AME2020)中2457个有精确值的原子核(Z≥8和N≥8)作为总数据集,随机选取其中80%的数据为训练集用于模型训练,通过预测余下的20%进行模型验证.最终两个数据集的误差均方根(RMS)基本相等,而且全部数据的RMS从2.9894 Me V降到0.5695 Me V,下降了80%,呈现出较好的结果.此模型进行了输入参数上的改进(区域限定策略),使得未知核(Z=118—126)可以被限定在一个固定的区域内,从而提高了预测的准确性.为了验证这一性质,对实验数据(Z=100—117)进行了预测计算,结果也与实验值符合得很好.最后使用该方案对未知元素Z=118—126进行了预测,为以后寻找新元素提供了新思路.
文摘A model of continuous-time insider trading in which a risk-neutral in-sider possesses two imperfect correlated signals of a risky asset is studied.By conditional expectation theory and filtering theory,we first establish three lemmas:normal corre-lation,equivalent pricing and equivalent profit,which can guarantee to turn our model into a model with insider knowing full information.Then we investigate the impact of the two correlated signals on the market equilibrium consisting of optimal insider trading strategy and semi-strong pricing rule.It shows that in the equilibrium,(1)the market depth is constant over time;(2)if the two noisy signals are not linerly correlated,then all private information of the insider is incorporated into prices in the end while the whole information on the asset value can not incorporated into prices in the end;(3)if the two noisy signals are linear correlated such that the insider can infer the whole information of the asset value,then our model turns into a model with insider knowing full information;(4)if the two noisy signals are the same then the total ex ant profit of the insider is increasing with the noise decreasing,while down to O as the noise going up to infinity;(5)if the two noisy signals are not linear correlated then with one noisy signal fixed,the total ex ante profit of the insider is single-peaked with a unique minimum with respect to the other noisy signal value,and furthermore as the noisy value going to O it gets its maximum,the profit in the case that the real value is observed.
