This paper presents an electrical impedance tomography(EIT)method using a partial-differential-equationconstrained optimization approach.The forward problem in the inversion framework is described by a complete electr...This paper presents an electrical impedance tomography(EIT)method using a partial-differential-equationconstrained optimization approach.The forward problem in the inversion framework is described by a complete electrodemodel(CEM),which seeks the electric potential within the domain and at surface electrodes considering the contact impedance between them.The finite element solution of the electric potential has been validated using a commercial code.The inverse medium problem for reconstructing the unknown electrical conductivity profile is formulated as an optimization problem constrained by the CEM.The method seeks the optimal solution of the domain’s electrical conductivity to minimize a Lagrangian functional consisting of a least-squares objective functional and a regularization term.Enforcing the stationarity of the Lagrangian leads to state,adjoint,and control problems,which constitute the Karush-Kuhn-Tucker(KKT)first-order optimality conditions.Subsequently,the electrical conductivity profile of the domain is iteratively updated by solving the KKT conditions in the reduced space of the control variable.Numerical results show that the relative error of the measured and calculated electric potentials after the inversion is less than 1%,demonstrating the successful reconstruction of heterogeneous electrical conductivity profiles using the proposed EIT method.This method thus represents an application framework for nondestructive evaluation of structures and geotechnical site characterization.展开更多
面向“30·60”双碳目标,矿区能源利用方式的绿色、经济、高效转型成为我国能源革命的迫切需求。西部矿区拥有丰富的可再生能源资源禀赋,但仍面临着可再生能源就地消纳困难,电力设备投资成本高、利用率低以及外送输电通道有限的困...面向“30·60”双碳目标,矿区能源利用方式的绿色、经济、高效转型成为我国能源革命的迫切需求。西部矿区拥有丰富的可再生能源资源禀赋,但仍面临着可再生能源就地消纳困难,电力设备投资成本高、利用率低以及外送输电通道有限的困难。为提升矿区用能清洁化程度,提升矿区能源供给的稳定性与可靠性,增强矿区对外部电网的支撑能力,提出全清洁能源下的高品质矿区能源系统(High-quality Coal Mine Energy System,HCMES)及其配置优化方法。首先,考虑西部矿山综合能源系统的负荷特点与伴生能源利用,结合可再生能源发电与废弃矿井抽水蓄能,构建全清洁能源下的HCMES架构。其次,考虑到矿区生产全流程负荷的需求响应能力,考虑系统的能量平衡约束,提出全清洁能源下的高品质矿区能源系统优化配置模型。最后,以系统年平均综合成本最小化为目标,将原问题转化为混合整数线性规划模型,求解生成高品质矿区能源系统优化配置方案。以我国西部某年产煤量1200万t的矿区实际数据为实例,验证所提模型与方法的有效性,并分析可再生能源出力与生产负荷需求不确定性对系统优化配置结果的影响。算例仿真设置了4种矿区能源系统配置方式:不配置储能、配置抽水蓄能、配置电化学储能、配置抽水蓄能(不外购电能)。结果表明,所提出的HCMES相较于其他配置方式可减少电气一次设备投资11.11%,相较于方式3可降低年平均综合成本7.91%,且最多可减少矿区生产用能总二氧化碳排放量91.17%。展开更多
近几年卷积神经网络作为深度学习最重要的技术,在图像分类、物体检测、语音识别等领域均有所建树。在此期间,由多层卷积神经网络组成的深度神经网络横空出世,在各种任务准确性方面具有显著提升。然而,神经网络的权重往往被限定在单精度...近几年卷积神经网络作为深度学习最重要的技术,在图像分类、物体检测、语音识别等领域均有所建树。在此期间,由多层卷积神经网络组成的深度神经网络横空出世,在各种任务准确性方面具有显著提升。然而,神经网络的权重往往被限定在单精度类型,使网络体积相较于特定硬件平台上的内存空间更大,且floating point 16、INT 8等单精度类型已无法满足现在一些模型推理的现实需求。为此,提出一种以子图为最小单位,通过判断相邻结点之间的融合关系,添加了丰富比特位的混合精度推理算法。首先,在原有单精度量化设计的搜索空间中增加floating point 16半精度的比特配置,使最终搜索空间变大,为寻找最优解提供更多机会。其次,使用子图融合的思想,通过整数线性规划将融合后的不同子图精度配置,根据模型大小、推理延迟和位宽操作数3个约束对计算图进行划分,使最后累积的扰动误差减少。最终,在ResNet系列网络上验证发现,所提模型精度相较于HAWQ V3的损失没超过1%的同时,相较于其他混合精度量化方法在推理速度方面得到了提升,在ResNet18网络中推理速度分别提升18.15%、19.21%,在ResNet50网络中推理速度分别提升13.15%、13.70%。展开更多
Flexible job shop scheduling problem(FJSP)is the core decision-making problem of intelligent manufacturing production management.The Harris hawk optimization(HHO)algorithm,as a typical metaheuristic algorithm,has been...Flexible job shop scheduling problem(FJSP)is the core decision-making problem of intelligent manufacturing production management.The Harris hawk optimization(HHO)algorithm,as a typical metaheuristic algorithm,has been widely employed to solve scheduling problems.However,HHO suffers from premature convergence when solving NP-hard problems.Therefore,this paper proposes an improved HHO algorithm(GNHHO)to solve the FJSP.GNHHO introduces an elitism strategy,a chaotic mechanism,a nonlinear escaping energy update strategy,and a Gaussian random walk strategy to prevent premature convergence.A flexible job shop scheduling model is constructed,and the static and dynamic FJSP is investigated to minimize the makespan.This paper chooses a two-segment encoding mode based on the job and the machine of the FJSP.To verify the effectiveness of GNHHO,this study tests it in 23 benchmark functions,10 standard job shop scheduling problems(JSPs),and 5 standard FJSPs.Besides,this study collects data from an agricultural company and uses the GNHHO algorithm to optimize the company’s FJSP.The optimized scheduling scheme demonstrates significant improvements in makespan,with an advancement of 28.16%for static scheduling and 35.63%for dynamic scheduling.Moreover,it achieves an average increase of 21.50%in the on-time order delivery rate.The results demonstrate that the performance of the GNHHO algorithm in solving FJSP is superior to some existing algorithms.展开更多
基金funded by the National Research Foundation of Korea,the Grant from a Basic Science and Engineering Research Project(NRF-2017R1C1B200497515)and the Grant from Basic Laboratory Support Project(NRF-2020R1A4A101882611).
文摘This paper presents an electrical impedance tomography(EIT)method using a partial-differential-equationconstrained optimization approach.The forward problem in the inversion framework is described by a complete electrodemodel(CEM),which seeks the electric potential within the domain and at surface electrodes considering the contact impedance between them.The finite element solution of the electric potential has been validated using a commercial code.The inverse medium problem for reconstructing the unknown electrical conductivity profile is formulated as an optimization problem constrained by the CEM.The method seeks the optimal solution of the domain’s electrical conductivity to minimize a Lagrangian functional consisting of a least-squares objective functional and a regularization term.Enforcing the stationarity of the Lagrangian leads to state,adjoint,and control problems,which constitute the Karush-Kuhn-Tucker(KKT)first-order optimality conditions.Subsequently,the electrical conductivity profile of the domain is iteratively updated by solving the KKT conditions in the reduced space of the control variable.Numerical results show that the relative error of the measured and calculated electric potentials after the inversion is less than 1%,demonstrating the successful reconstruction of heterogeneous electrical conductivity profiles using the proposed EIT method.This method thus represents an application framework for nondestructive evaluation of structures and geotechnical site characterization.
文摘面向“30·60”双碳目标,矿区能源利用方式的绿色、经济、高效转型成为我国能源革命的迫切需求。西部矿区拥有丰富的可再生能源资源禀赋,但仍面临着可再生能源就地消纳困难,电力设备投资成本高、利用率低以及外送输电通道有限的困难。为提升矿区用能清洁化程度,提升矿区能源供给的稳定性与可靠性,增强矿区对外部电网的支撑能力,提出全清洁能源下的高品质矿区能源系统(High-quality Coal Mine Energy System,HCMES)及其配置优化方法。首先,考虑西部矿山综合能源系统的负荷特点与伴生能源利用,结合可再生能源发电与废弃矿井抽水蓄能,构建全清洁能源下的HCMES架构。其次,考虑到矿区生产全流程负荷的需求响应能力,考虑系统的能量平衡约束,提出全清洁能源下的高品质矿区能源系统优化配置模型。最后,以系统年平均综合成本最小化为目标,将原问题转化为混合整数线性规划模型,求解生成高品质矿区能源系统优化配置方案。以我国西部某年产煤量1200万t的矿区实际数据为实例,验证所提模型与方法的有效性,并分析可再生能源出力与生产负荷需求不确定性对系统优化配置结果的影响。算例仿真设置了4种矿区能源系统配置方式:不配置储能、配置抽水蓄能、配置电化学储能、配置抽水蓄能(不外购电能)。结果表明,所提出的HCMES相较于其他配置方式可减少电气一次设备投资11.11%,相较于方式3可降低年平均综合成本7.91%,且最多可减少矿区生产用能总二氧化碳排放量91.17%。
文摘近几年卷积神经网络作为深度学习最重要的技术,在图像分类、物体检测、语音识别等领域均有所建树。在此期间,由多层卷积神经网络组成的深度神经网络横空出世,在各种任务准确性方面具有显著提升。然而,神经网络的权重往往被限定在单精度类型,使网络体积相较于特定硬件平台上的内存空间更大,且floating point 16、INT 8等单精度类型已无法满足现在一些模型推理的现实需求。为此,提出一种以子图为最小单位,通过判断相邻结点之间的融合关系,添加了丰富比特位的混合精度推理算法。首先,在原有单精度量化设计的搜索空间中增加floating point 16半精度的比特配置,使最终搜索空间变大,为寻找最优解提供更多机会。其次,使用子图融合的思想,通过整数线性规划将融合后的不同子图精度配置,根据模型大小、推理延迟和位宽操作数3个约束对计算图进行划分,使最后累积的扰动误差减少。最终,在ResNet系列网络上验证发现,所提模型精度相较于HAWQ V3的损失没超过1%的同时,相较于其他混合精度量化方法在推理速度方面得到了提升,在ResNet18网络中推理速度分别提升18.15%、19.21%,在ResNet50网络中推理速度分别提升13.15%、13.70%。
文摘Flexible job shop scheduling problem(FJSP)is the core decision-making problem of intelligent manufacturing production management.The Harris hawk optimization(HHO)algorithm,as a typical metaheuristic algorithm,has been widely employed to solve scheduling problems.However,HHO suffers from premature convergence when solving NP-hard problems.Therefore,this paper proposes an improved HHO algorithm(GNHHO)to solve the FJSP.GNHHO introduces an elitism strategy,a chaotic mechanism,a nonlinear escaping energy update strategy,and a Gaussian random walk strategy to prevent premature convergence.A flexible job shop scheduling model is constructed,and the static and dynamic FJSP is investigated to minimize the makespan.This paper chooses a two-segment encoding mode based on the job and the machine of the FJSP.To verify the effectiveness of GNHHO,this study tests it in 23 benchmark functions,10 standard job shop scheduling problems(JSPs),and 5 standard FJSPs.Besides,this study collects data from an agricultural company and uses the GNHHO algorithm to optimize the company’s FJSP.The optimized scheduling scheme demonstrates significant improvements in makespan,with an advancement of 28.16%for static scheduling and 35.63%for dynamic scheduling.Moreover,it achieves an average increase of 21.50%in the on-time order delivery rate.The results demonstrate that the performance of the GNHHO algorithm in solving FJSP is superior to some existing algorithms.
