The precise integration method proposed for linear time-invariant homogeneous dynamic systems can provide accurate numerical results that approach an exact solution at integration points. However, difficulties arise w...The precise integration method proposed for linear time-invariant homogeneous dynamic systems can provide accurate numerical results that approach an exact solution at integration points. However, difficulties arise when the algorithm is used for non-homogeneous dynamic systems due to the inverse matrix calculation required. In this paper, the structural dynamic equalibrium equations are converted into a special form, the inverse matrix calculation is replaced by the Crout decomposition method to solve the dynamic equilibrium equations, and the precise integration method without the inverse matrix calculation is obtained. The new algorithm enhances the present precise integration method by improving both the computational accuracy and efficiency. Two numerical examples are given to demonstrate the validity and efficiency of the proposed algorithm.展开更多
In this paper,we propose genetic programming(GP) using dynamic population variation(DPV) with four innovations for reducing computational efforts.A new stagnation phase definition and characteristic measure are define...In this paper,we propose genetic programming(GP) using dynamic population variation(DPV) with four innovations for reducing computational efforts.A new stagnation phase definition and characteristic measure are defined for our DPV.The exponential pivot function is proposed to our DPV method in conjunction with the new stagnation phase definition.An appropriate population variation formula is suggested to accelerate convergence.The efficacy of these innovations in our DPV is examined using six benchmark problems.Comparison among the difierent characteristic measures has been conducted for regression problems and the new proposed measure outperformed other measures.It is proved that our DPV has the capacity to provide solutions at a lower computational effort compared with previously proposed DPV methods and standard genetic programming in most cases.Meanwhile,our DPV approach introduced in GP could also rapidly find an excellent solution as well as standard GP in system modeling problems.展开更多
针对超临界锅炉运行中氧化膜动态生长特性难以实时监测导致的机组运行效能劣化问题,本文提出一种高效融合深度学习与生长机理的氧化膜厚度预测模型(VMD-NPDCLO-BiLSTM-SAMME),利用变分模态分解(variational mode decomposition,VMD)将...针对超临界锅炉运行中氧化膜动态生长特性难以实时监测导致的机组运行效能劣化问题,本文提出一种高效融合深度学习与生长机理的氧化膜厚度预测模型(VMD-NPDCLO-BiLSTM-SAMME),利用变分模态分解(variational mode decomposition,VMD)将原始序列分解为若干相对平稳的子序列,采用混沌-莱维神经种群动态优化(neural population dynamics with chaotic-levy optimization,NPDCLO)算法,构建具有最优超参数配置的双向长短期记忆神经网络(bidirectional long short-term memory,BiLSTM)模型,并使用多类指数损失函数渐进添加模型(stagewise additive modeling using a multi-class exponential loss function,SAMME)将多个NPDCLO-BiLSTM弱分类器组合,构建VMD-NPDCLO-BiLSTM-SAMME强分类器模型对氧化膜生成机理模型中的高温过热器壁温与烟温进行预测,最终利用预测结果融合机理模型以实现氧化膜厚度的实时精确估计。仿真实验结果表明:本文提出的模型与现有的BiLSTM-SAMME模型相比,壁温的平均绝对误差与均方根误差分别降低32.52%、32.26%,烟温的平均绝对误差与均方根误差分别降低47.38%、55.27%;氧化膜厚度预测模型的平均误差为7.42%,验证了模型的有效性及工程适用性。展开更多
文摘The precise integration method proposed for linear time-invariant homogeneous dynamic systems can provide accurate numerical results that approach an exact solution at integration points. However, difficulties arise when the algorithm is used for non-homogeneous dynamic systems due to the inverse matrix calculation required. In this paper, the structural dynamic equalibrium equations are converted into a special form, the inverse matrix calculation is replaced by the Crout decomposition method to solve the dynamic equilibrium equations, and the precise integration method without the inverse matrix calculation is obtained. The new algorithm enhances the present precise integration method by improving both the computational accuracy and efficiency. Two numerical examples are given to demonstrate the validity and efficiency of the proposed algorithm.
基金Ministry of Major Science & Technology of Shanghai(No.10DZ1200204)
文摘In this paper,we propose genetic programming(GP) using dynamic population variation(DPV) with four innovations for reducing computational efforts.A new stagnation phase definition and characteristic measure are defined for our DPV.The exponential pivot function is proposed to our DPV method in conjunction with the new stagnation phase definition.An appropriate population variation formula is suggested to accelerate convergence.The efficacy of these innovations in our DPV is examined using six benchmark problems.Comparison among the difierent characteristic measures has been conducted for regression problems and the new proposed measure outperformed other measures.It is proved that our DPV has the capacity to provide solutions at a lower computational effort compared with previously proposed DPV methods and standard genetic programming in most cases.Meanwhile,our DPV approach introduced in GP could also rapidly find an excellent solution as well as standard GP in system modeling problems.
文摘针对超临界锅炉运行中氧化膜动态生长特性难以实时监测导致的机组运行效能劣化问题,本文提出一种高效融合深度学习与生长机理的氧化膜厚度预测模型(VMD-NPDCLO-BiLSTM-SAMME),利用变分模态分解(variational mode decomposition,VMD)将原始序列分解为若干相对平稳的子序列,采用混沌-莱维神经种群动态优化(neural population dynamics with chaotic-levy optimization,NPDCLO)算法,构建具有最优超参数配置的双向长短期记忆神经网络(bidirectional long short-term memory,BiLSTM)模型,并使用多类指数损失函数渐进添加模型(stagewise additive modeling using a multi-class exponential loss function,SAMME)将多个NPDCLO-BiLSTM弱分类器组合,构建VMD-NPDCLO-BiLSTM-SAMME强分类器模型对氧化膜生成机理模型中的高温过热器壁温与烟温进行预测,最终利用预测结果融合机理模型以实现氧化膜厚度的实时精确估计。仿真实验结果表明:本文提出的模型与现有的BiLSTM-SAMME模型相比,壁温的平均绝对误差与均方根误差分别降低32.52%、32.26%,烟温的平均绝对误差与均方根误差分别降低47.38%、55.27%;氧化膜厚度预测模型的平均误差为7.42%,验证了模型的有效性及工程适用性。
