In this work,a method is put forward to obtain the dynamic solution efficiently and accurately for a large-scale train-track-substructure(TTS)system.It is called implicit-explicit integration and multi-time-step solut...In this work,a method is put forward to obtain the dynamic solution efficiently and accurately for a large-scale train-track-substructure(TTS)system.It is called implicit-explicit integration and multi-time-step solution method(abbreviated as mI-nE-MTS method).The TTS system is divided into train-track subsystem and substruc-ture subsystem.Considering that the root cause of low effi-ciency of obtaining TTS solution lies in solving the alge-braic equation of the substructures,the high-efficient Zhai method,an explicit integration scheme,can be introduced to avoid matrix inversion process.The train-track system is solved by implicitly Park method.Moreover,it is known that the requirement of time step size differs for different sub-systems,integration methods and structural frequency response characteristics.A multi-time-step solution is pro-posed,in which time step size for the train-track subsystem and the substructure subsystem can be arbitrarily chosen once satisfying stability and precision demand,namely the time spent for m implicit integral steps is equal to n explicit integral steps,i.e.,mI=nE as mentioned above.The numeri-cal examples show the accuracy,efficiency,and engineering practicality of the proposed method.展开更多
对水质情况进行准确评估和预测对水污染防控至关重要,然而,由于水质受多种因素的影响,其时间序列数据常常具有趋势性、季节性和长期依赖关系,传统的预测方法往往无法很好地捕捉这些特征。为了解决这些问题,首先基于STL(Seasonal and Tre...对水质情况进行准确评估和预测对水污染防控至关重要,然而,由于水质受多种因素的影响,其时间序列数据常常具有趋势性、季节性和长期依赖关系,传统的预测方法往往无法很好地捕捉这些特征。为了解决这些问题,首先基于STL(Seasonal and Trend Decomposition using Loess)和TCN(Temporal Convolutional Network)构建STL-TCN水质预测模型。其中,通过STL模型对水质时间序列数据进行趋势和季节性分解,有效地提取时序数据的周期性特征;利用TCN模型中并行结构和残差连接有效捕捉时间序列数据的长期依赖关系,对分解后的数据进行多步预测。然后,选用福建省浪石断面河流的氨氮数据来验证STL-TCN水质预测模型的预测效果,并与基于长短时记忆网络(LSTM)和门控循环单元结构(GRU)的水质预测模型进行对比实验。实验结果表明,STL-TCN水质预测模型12步预测的MAE平均值达到0.0343、RMSE平均值达到0.0494、R^(2)平均值达到0.94737,相对LSTM和GRU,MAE平均提高7.8%和8.1%、RMSE平均提高2.2%和1.8%、R^(2)平均提高7.9%和7.8%。说明STL-TCN水质预测模型能够有效提高水质预测的准确性和稳定性,可以作为辅助水环境管理和决策的一种有效手段。展开更多
针对复杂的应用环境下,时间序列建模不易准确,多步预测精度不高的问题,提出基于粒子滤波(particle filter,PF)优化的滚动式时间序列(roll time series,RTS)多步预测算法(PF_RTS)。采用Box-Jenkins方法对时间序列滚动自适应建模,所建模...针对复杂的应用环境下,时间序列建模不易准确,多步预测精度不高的问题,提出基于粒子滤波(particle filter,PF)优化的滚动式时间序列(roll time series,RTS)多步预测算法(PF_RTS)。采用Box-Jenkins方法对时间序列滚动自适应建模,所建模型作为粒子的状态转移方程,利用粒子滤波算法实时动态修正预测数据,逼近状态的最优估计。本文算法具有自学习能力,适合实时应用。仿真结果表明,本文算法需要的先验知识少,提高了预测的精度。展开更多
基金This work was supported by the National Natural Science Foundation of China(Grant Nos.52008404,U1934217 and 11790283)Science and Technology Research and Development Program Project of China Railway Group Limited(Major Special Project,No.2020-Special-02)the National Natural Science Foundation of Hunan Province(Grant No.2021JJ30850).
文摘In this work,a method is put forward to obtain the dynamic solution efficiently and accurately for a large-scale train-track-substructure(TTS)system.It is called implicit-explicit integration and multi-time-step solution method(abbreviated as mI-nE-MTS method).The TTS system is divided into train-track subsystem and substruc-ture subsystem.Considering that the root cause of low effi-ciency of obtaining TTS solution lies in solving the alge-braic equation of the substructures,the high-efficient Zhai method,an explicit integration scheme,can be introduced to avoid matrix inversion process.The train-track system is solved by implicitly Park method.Moreover,it is known that the requirement of time step size differs for different sub-systems,integration methods and structural frequency response characteristics.A multi-time-step solution is pro-posed,in which time step size for the train-track subsystem and the substructure subsystem can be arbitrarily chosen once satisfying stability and precision demand,namely the time spent for m implicit integral steps is equal to n explicit integral steps,i.e.,mI=nE as mentioned above.The numeri-cal examples show the accuracy,efficiency,and engineering practicality of the proposed method.
文摘对水质情况进行准确评估和预测对水污染防控至关重要,然而,由于水质受多种因素的影响,其时间序列数据常常具有趋势性、季节性和长期依赖关系,传统的预测方法往往无法很好地捕捉这些特征。为了解决这些问题,首先基于STL(Seasonal and Trend Decomposition using Loess)和TCN(Temporal Convolutional Network)构建STL-TCN水质预测模型。其中,通过STL模型对水质时间序列数据进行趋势和季节性分解,有效地提取时序数据的周期性特征;利用TCN模型中并行结构和残差连接有效捕捉时间序列数据的长期依赖关系,对分解后的数据进行多步预测。然后,选用福建省浪石断面河流的氨氮数据来验证STL-TCN水质预测模型的预测效果,并与基于长短时记忆网络(LSTM)和门控循环单元结构(GRU)的水质预测模型进行对比实验。实验结果表明,STL-TCN水质预测模型12步预测的MAE平均值达到0.0343、RMSE平均值达到0.0494、R^(2)平均值达到0.94737,相对LSTM和GRU,MAE平均提高7.8%和8.1%、RMSE平均提高2.2%和1.8%、R^(2)平均提高7.9%和7.8%。说明STL-TCN水质预测模型能够有效提高水质预测的准确性和稳定性,可以作为辅助水环境管理和决策的一种有效手段。
文摘针对复杂的应用环境下,时间序列建模不易准确,多步预测精度不高的问题,提出基于粒子滤波(particle filter,PF)优化的滚动式时间序列(roll time series,RTS)多步预测算法(PF_RTS)。采用Box-Jenkins方法对时间序列滚动自适应建模,所建模型作为粒子的状态转移方程,利用粒子滤波算法实时动态修正预测数据,逼近状态的最优估计。本文算法具有自学习能力,适合实时应用。仿真结果表明,本文算法需要的先验知识少,提高了预测的精度。
