Randomness and fuzziness are among the attributes of the influential factors for stability assessment of pile foundation. According to these two characteristics, the triangular fuzzy number analysis approach was intro...Randomness and fuzziness are among the attributes of the influential factors for stability assessment of pile foundation. According to these two characteristics, the triangular fuzzy number analysis approach was introduced to determine the probability-distributed function of mechanical parameters. Then the functional function of reliability analysis was constructed based on the study of bearing mechanism of pile foundation, and the way to calculate interval values of the functional function was developed by using improved interval-truncation approach and operation rules of interval numbers. Afterwards, the non-probabilistic fuzzy reliability analysis method was applied to assessing the pile foundation, from which a method was presented for non- probabilistic fuzzy reliability analysis of pile foundation stability by interval theory. Finally, the probability distribution curve of non- probabilistic fuzzy reliability indexes of practical pile foundation was concluded. Its failure possibility is 0.91%, which shows that the pile foundation is stable and reliable.展开更多
A robust reliability method for stability analysis and reliability-based stabilization of time-delay dynamic systems with uncertain but bounded parameters is presented by treating the uncertain parameters as interval ...A robust reliability method for stability analysis and reliability-based stabilization of time-delay dynamic systems with uncertain but bounded parameters is presented by treating the uncertain parameters as interval variables.The performance function used for robust reliability analysis is defined by a delayindependent stability criterion.The design of robust controllers is carried out by solving a reliability-based optimization problem in which the control cost satisfying design requirements is minimized.This kind of treatment makes it possible to achieve a balance between the reliability and control cost in the design of controller when uncertainties must be taken into account.By the method,a robust reliability measure of the degree of stability of a time-delay uncertain system can be provided,and the maximum robustness bounds of uncertain parameters such that the time-delay system to be stable can be obtained.All the procedures are based on the linear matrix inequality approach and therefore can be carried out conveniently.The effectiveness and feasibility of the proposed method are demonstrated with two practical examples.It is shown by numerical simulations and comparison that it is meaningful to take the robust reliability into account in the control design of uncertain systems.展开更多
The robust reliable H∞ control problem for discrete-time Markovian jump systems with actuator failures is studied. A more practical model of actuator failures than outage is considered. Based on the state feedback me...The robust reliable H∞ control problem for discrete-time Markovian jump systems with actuator failures is studied. A more practical model of actuator failures than outage is considered. Based on the state feedback method, the resulting closed-loop systems are reliable in that they remain robust stochastically stable and satisfy a certain level of H∞ disturbance attenuation not only when all actuators are operational, but also in case of some actuator failures, The solvability condition of controllers can be equivalent to a feasibility problem of coupled linear matrix inequalities （LMIs）. A numerical example is also given to illustrate the design procedures and their effectiveness.展开更多
随着可再生能源占比的提升,气电联合配网(integrated electricity-gas distribution systems,IEGDS)发展迅速,其研究的必要性也随之凸显。为解决IEGDS考虑可再生能源接入后的多设施规划及可靠性评估问题,提出了考虑可靠性约束的IEGDS分...随着可再生能源占比的提升,气电联合配网(integrated electricity-gas distribution systems,IEGDS)发展迅速,其研究的必要性也随之凸显。为解决IEGDS考虑可再生能源接入后的多设施规划及可靠性评估问题,提出了考虑可靠性约束的IEGDS分布鲁棒优化(distributionally robust optimization,DRO)规划模型。可靠性评估通过系统平均中断频率指标(system average interruption frequency index,SAIFI)与预期能源未供应指标(expectation energy not supplied,EENS)量化,给出一种显式可靠性评估模型。利用Wasserstein距离衡量的DRO规划模型来处理可再生能源出力、电与气负荷的随机性。最后经过案例分析得出,投建设施种类的增加不仅能降低总成本,还能提高系统运行可靠性与灵活性,可再生能源出力与气电负荷的不确定性导致总成本提升60%以上。随着可靠性要求提升总成本也随之提高,但得到的方案在实际应用中更具价值与参考性。展开更多
大语言模型(large language models,LLMs)所展现的处理各种自然语言处理(natural language processing,NLP)任务的能力引发了广泛关注.然而,它们在处理现实中各种复杂场景时的鲁棒性尚未得到充分探索,这对于评估模型的稳定性和可靠性尤...大语言模型(large language models,LLMs)所展现的处理各种自然语言处理(natural language processing,NLP)任务的能力引发了广泛关注.然而,它们在处理现实中各种复杂场景时的鲁棒性尚未得到充分探索,这对于评估模型的稳定性和可靠性尤为重要.因此,使用涵盖了9个常见NLP任务的15个数据集(约147000个原始测试样本)和来自TextFlint的61种鲁棒的文本变形方法分析GPT-3和GPT-3.5系列模型在原始数据集上的性能,以及其在不同任务和文本变形级别(字符、词和句子)上的鲁棒性.研究结果表明,GPT模型虽然在情感分析、语义匹配等分类任务和阅读理解任务中表现出良好的性能,但其处理信息抽取任务的能力仍较为欠缺,比如其对关系抽取任务中各种关系类型存在严重混淆,甚至出现“幻觉”现象.在鲁棒性评估实验中,GPT模型在任务层面和变形层面的鲁棒性都较弱,其中,在分类任务和句子级别的变形中鲁棒性缺乏更为显著.此外,探究了模型迭代过程中性能和鲁棒性的变化,以及提示中的演示数量和演示内容对模型性能和鲁棒性的影响.结果表明,随着模型的迭代以及上下文学习的加入,模型的性能稳步提升,但是鲁棒性依然亟待提升.这些发现从任务类型、变形种类、提示内容等方面揭示了GPT模型还无法完全胜任常见的NLP任务,并且模型存在的鲁棒性问题难以通过提升模型性能或改变提示内容等方式解决.通过对gpt-3.5-turbo的更新版本、gpt-4模型,以及开源模型LLaMA2-7B和LLaMA2-13B的性能和鲁棒性表现进行对比,进一步验证了实验结论.鉴于此,未来的大模型研究应当提升模型在信息提取以及语义理解等方面的能力,并且应当在模型训练或微调阶段考虑提升其鲁棒性.展开更多
基金Project(50378036) supported by the National Natural Science Foundation of ChinaProject(03JJY5024) supported by the Natural Science Foundation of Hunan Province, China
文摘Randomness and fuzziness are among the attributes of the influential factors for stability assessment of pile foundation. According to these two characteristics, the triangular fuzzy number analysis approach was introduced to determine the probability-distributed function of mechanical parameters. Then the functional function of reliability analysis was constructed based on the study of bearing mechanism of pile foundation, and the way to calculate interval values of the functional function was developed by using improved interval-truncation approach and operation rules of interval numbers. Afterwards, the non-probabilistic fuzzy reliability analysis method was applied to assessing the pile foundation, from which a method was presented for non- probabilistic fuzzy reliability analysis of pile foundation stability by interval theory. Finally, the probability distribution curve of non- probabilistic fuzzy reliability indexes of practical pile foundation was concluded. Its failure possibility is 0.91%, which shows that the pile foundation is stable and reliable.
文摘A robust reliability method for stability analysis and reliability-based stabilization of time-delay dynamic systems with uncertain but bounded parameters is presented by treating the uncertain parameters as interval variables.The performance function used for robust reliability analysis is defined by a delayindependent stability criterion.The design of robust controllers is carried out by solving a reliability-based optimization problem in which the control cost satisfying design requirements is minimized.This kind of treatment makes it possible to achieve a balance between the reliability and control cost in the design of controller when uncertainties must be taken into account.By the method,a robust reliability measure of the degree of stability of a time-delay uncertain system can be provided,and the maximum robustness bounds of uncertain parameters such that the time-delay system to be stable can be obtained.All the procedures are based on the linear matrix inequality approach and therefore can be carried out conveniently.The effectiveness and feasibility of the proposed method are demonstrated with two practical examples.It is shown by numerical simulations and comparison that it is meaningful to take the robust reliability into account in the control design of uncertain systems.
基金the National Natural Science Foundation of China (60574001)Program for New Century Excellent Talents in University (05-0485)Program for Innovative Research Team of Jiangnan University
文摘The robust reliable H∞ control problem for discrete-time Markovian jump systems with actuator failures is studied. A more practical model of actuator failures than outage is considered. Based on the state feedback method, the resulting closed-loop systems are reliable in that they remain robust stochastically stable and satisfy a certain level of H∞ disturbance attenuation not only when all actuators are operational, but also in case of some actuator failures, The solvability condition of controllers can be equivalent to a feasibility problem of coupled linear matrix inequalities （LMIs）. A numerical example is also given to illustrate the design procedures and their effectiveness.
文摘随着可再生能源占比的提升,气电联合配网(integrated electricity-gas distribution systems,IEGDS)发展迅速,其研究的必要性也随之凸显。为解决IEGDS考虑可再生能源接入后的多设施规划及可靠性评估问题,提出了考虑可靠性约束的IEGDS分布鲁棒优化(distributionally robust optimization,DRO)规划模型。可靠性评估通过系统平均中断频率指标(system average interruption frequency index,SAIFI)与预期能源未供应指标(expectation energy not supplied,EENS)量化,给出一种显式可靠性评估模型。利用Wasserstein距离衡量的DRO规划模型来处理可再生能源出力、电与气负荷的随机性。最后经过案例分析得出,投建设施种类的增加不仅能降低总成本,还能提高系统运行可靠性与灵活性,可再生能源出力与气电负荷的不确定性导致总成本提升60%以上。随着可靠性要求提升总成本也随之提高,但得到的方案在实际应用中更具价值与参考性。
文摘大语言模型(large language models,LLMs)所展现的处理各种自然语言处理(natural language processing,NLP)任务的能力引发了广泛关注.然而,它们在处理现实中各种复杂场景时的鲁棒性尚未得到充分探索,这对于评估模型的稳定性和可靠性尤为重要.因此,使用涵盖了9个常见NLP任务的15个数据集(约147000个原始测试样本)和来自TextFlint的61种鲁棒的文本变形方法分析GPT-3和GPT-3.5系列模型在原始数据集上的性能,以及其在不同任务和文本变形级别(字符、词和句子)上的鲁棒性.研究结果表明,GPT模型虽然在情感分析、语义匹配等分类任务和阅读理解任务中表现出良好的性能,但其处理信息抽取任务的能力仍较为欠缺,比如其对关系抽取任务中各种关系类型存在严重混淆,甚至出现“幻觉”现象.在鲁棒性评估实验中,GPT模型在任务层面和变形层面的鲁棒性都较弱,其中,在分类任务和句子级别的变形中鲁棒性缺乏更为显著.此外,探究了模型迭代过程中性能和鲁棒性的变化,以及提示中的演示数量和演示内容对模型性能和鲁棒性的影响.结果表明,随着模型的迭代以及上下文学习的加入,模型的性能稳步提升,但是鲁棒性依然亟待提升.这些发现从任务类型、变形种类、提示内容等方面揭示了GPT模型还无法完全胜任常见的NLP任务,并且模型存在的鲁棒性问题难以通过提升模型性能或改变提示内容等方式解决.通过对gpt-3.5-turbo的更新版本、gpt-4模型,以及开源模型LLaMA2-7B和LLaMA2-13B的性能和鲁棒性表现进行对比,进一步验证了实验结论.鉴于此,未来的大模型研究应当提升模型在信息提取以及语义理解等方面的能力,并且应当在模型训练或微调阶段考虑提升其鲁棒性.
