联邦学习是一种新型的分布式机器学习方法,可以在不共享原始数据的前提下训练模型。当前,联邦学习方法存在针对模型准确率最优化、通信成本最优化、参与者性能分布均衡等多个目标同时优化难的问题,难以做到多目标的同步均衡。针对该问题...联邦学习是一种新型的分布式机器学习方法,可以在不共享原始数据的前提下训练模型。当前,联邦学习方法存在针对模型准确率最优化、通信成本最优化、参与者性能分布均衡等多个目标同时优化难的问题,难以做到多目标的同步均衡。针对该问题,提出联邦学习四目标优化模型及求解算法。将全局模型错误率、模型准确率分布方差、通信成本、数据成本作为优化目标,构建优化模型。同时,针对该模型的求解搜索空间大,传统NSGA-Ⅲ算法难以寻优的问题,提出基于佳点集初始化策略的改进NSGA-Ⅲ联邦学习多目标优化算法GPNSGA-Ⅲ(Good Point Set Initialization NSGA-Ⅲ),以求取Pareto最优解。该算法通过佳点集初始化策略将有限的初始化种群以均匀的方式分布在目标求解空间中,相较于原始算法,使第一代解最大限度地接近最优值,提升寻优能力。实验结果证明,GPNSGA-Ⅲ算法得到的Pareto解的超体积值相较于NSGA-Ⅲ算法平均提升107%;Spacing值相较于NSGA-Ⅲ算法平均下降32.3%;对比其他多目标优化算法,GPNSGA-Ⅲ算法能在保证模型准确率的情况下,更有效地实现模型分布方差、通信成本和数据成本的均衡。展开更多
半球谐振陀螺是目前精度最高的一种振动陀螺。对于半球谐振陀螺,在制造过程中难以完全避免的工艺缺陷会导致谐振子的质量、刚度、品质因数、密度、弹性模量、阻尼等参数周向分布不均匀,产生频率裂解现象,使得主、次振动存在误差耦合。...半球谐振陀螺是目前精度最高的一种振动陀螺。对于半球谐振陀螺,在制造过程中难以完全避免的工艺缺陷会导致谐振子的质量、刚度、品质因数、密度、弹性模量、阻尼等参数周向分布不均匀,产生频率裂解现象,使得主、次振动存在误差耦合。传统的频率分裂补偿方法会导致半球谐振陀螺的品质因数降低,且存在补偿成本高、操作复杂等问题。提出了一种电平衡补偿方案,通过对不同的电极施加静电力改变谐振子的刚度,实现对频率裂解的补偿,并结合NSGA-Ⅲ多目标优化算法对补偿参数进行了优化,首次将静电修调方案对谐振子本身性能的影响、功耗问题以及该方案所能提供的频率裂解补偿值同时进行考虑,以实现针对半球谐振陀螺频率裂解的最优补偿。经过验证,该方法针对不同的谐振子和频率裂解在所选参数下能够给出最优的补偿方案,频率分裂补偿值提高了50.2%,补偿电压的需求分别降低6.3%和56.3%,补偿精度高于0.5 m Hz;补偿后谐振子的检测误差降低了一个数量级,固有频率仅降低2.3%。该方案可以有效提高陀螺仪的动态性能,为半球谐振陀螺的频率裂解的最优化补偿提供了一种参考,且该方法同样适用于杯型、环形等哥氏陀螺仪。展开更多
Finding an optimal isolator arrangement for asymmetric structures using traditional conceptual design methods that can significantly minimize torsional response while ensuring efficient horizontal seismic isolation is...Finding an optimal isolator arrangement for asymmetric structures using traditional conceptual design methods that can significantly minimize torsional response while ensuring efficient horizontal seismic isolation is cumbersome and inefficient.Thus,this work develops a multi-objective optimization method to enhance the torsional resistance of asymmetric base-isolated structures.The primary objective is to simultaneously minimize the interstory rotation of the superstructure,the rotation of the isolation layer,and the interstory displacement of the superstructure without exceeding the isolator displacement limits.A fast non-dominated sorting genetic algorithm(NSGA-Ⅱ)is employed to satisfy this optimization objective.Subsequently,the isolator arrangement,encompassing both positions and categories,is optimized according to this multi-objective optimization method.Additionally,an optimization design platform is developed to streamline the design operation.This platform integrates the input of optimization parameters,the output of optimization results,the finite element analysis,and the multi-objective optimization method proposed herein.Finally,the application of this multi-objective optimization method and its associated platform are demonstrated on two asymmetric base-isolated structures of varying heights and plan configurations.The results indicate that the optimal isolator arrangement derived from the optimization method can further improve the control over the lateral and torsional responses of asymmetric base-isolated structures compared to conventional conceptual design methods.Notably,the interstory rotation of the optimal base-isolated structure is significantly reduced,constituting only approximately 33.7%of that observed in the original base-isolated structure.The proposed platform facilitates the automatic generation of the optimal design scheme for the isolators of asymmetric base-isolated structures,offering valuable insights and guidance for the burgeoning field of intelligent civil engineering design.展开更多
文摘联邦学习是一种新型的分布式机器学习方法,可以在不共享原始数据的前提下训练模型。当前,联邦学习方法存在针对模型准确率最优化、通信成本最优化、参与者性能分布均衡等多个目标同时优化难的问题,难以做到多目标的同步均衡。针对该问题,提出联邦学习四目标优化模型及求解算法。将全局模型错误率、模型准确率分布方差、通信成本、数据成本作为优化目标,构建优化模型。同时,针对该模型的求解搜索空间大,传统NSGA-Ⅲ算法难以寻优的问题,提出基于佳点集初始化策略的改进NSGA-Ⅲ联邦学习多目标优化算法GPNSGA-Ⅲ(Good Point Set Initialization NSGA-Ⅲ),以求取Pareto最优解。该算法通过佳点集初始化策略将有限的初始化种群以均匀的方式分布在目标求解空间中,相较于原始算法,使第一代解最大限度地接近最优值,提升寻优能力。实验结果证明,GPNSGA-Ⅲ算法得到的Pareto解的超体积值相较于NSGA-Ⅲ算法平均提升107%;Spacing值相较于NSGA-Ⅲ算法平均下降32.3%;对比其他多目标优化算法,GPNSGA-Ⅲ算法能在保证模型准确率的情况下,更有效地实现模型分布方差、通信成本和数据成本的均衡。
文摘半球谐振陀螺是目前精度最高的一种振动陀螺。对于半球谐振陀螺,在制造过程中难以完全避免的工艺缺陷会导致谐振子的质量、刚度、品质因数、密度、弹性模量、阻尼等参数周向分布不均匀,产生频率裂解现象,使得主、次振动存在误差耦合。传统的频率分裂补偿方法会导致半球谐振陀螺的品质因数降低,且存在补偿成本高、操作复杂等问题。提出了一种电平衡补偿方案,通过对不同的电极施加静电力改变谐振子的刚度,实现对频率裂解的补偿,并结合NSGA-Ⅲ多目标优化算法对补偿参数进行了优化,首次将静电修调方案对谐振子本身性能的影响、功耗问题以及该方案所能提供的频率裂解补偿值同时进行考虑,以实现针对半球谐振陀螺频率裂解的最优补偿。经过验证,该方法针对不同的谐振子和频率裂解在所选参数下能够给出最优的补偿方案,频率分裂补偿值提高了50.2%,补偿电压的需求分别降低6.3%和56.3%,补偿精度高于0.5 m Hz;补偿后谐振子的检测误差降低了一个数量级,固有频率仅降低2.3%。该方案可以有效提高陀螺仪的动态性能,为半球谐振陀螺的频率裂解的最优化补偿提供了一种参考,且该方法同样适用于杯型、环形等哥氏陀螺仪。
基金National Natural Science Foundation of China under Grant No.52278490。
文摘Finding an optimal isolator arrangement for asymmetric structures using traditional conceptual design methods that can significantly minimize torsional response while ensuring efficient horizontal seismic isolation is cumbersome and inefficient.Thus,this work develops a multi-objective optimization method to enhance the torsional resistance of asymmetric base-isolated structures.The primary objective is to simultaneously minimize the interstory rotation of the superstructure,the rotation of the isolation layer,and the interstory displacement of the superstructure without exceeding the isolator displacement limits.A fast non-dominated sorting genetic algorithm(NSGA-Ⅱ)is employed to satisfy this optimization objective.Subsequently,the isolator arrangement,encompassing both positions and categories,is optimized according to this multi-objective optimization method.Additionally,an optimization design platform is developed to streamline the design operation.This platform integrates the input of optimization parameters,the output of optimization results,the finite element analysis,and the multi-objective optimization method proposed herein.Finally,the application of this multi-objective optimization method and its associated platform are demonstrated on two asymmetric base-isolated structures of varying heights and plan configurations.The results indicate that the optimal isolator arrangement derived from the optimization method can further improve the control over the lateral and torsional responses of asymmetric base-isolated structures compared to conventional conceptual design methods.Notably,the interstory rotation of the optimal base-isolated structure is significantly reduced,constituting only approximately 33.7%of that observed in the original base-isolated structure.The proposed platform facilitates the automatic generation of the optimal design scheme for the isolators of asymmetric base-isolated structures,offering valuable insights and guidance for the burgeoning field of intelligent civil engineering design.
