Since in practical industrial applications, soft sensors based on static models often lead to low accuracy and poor robustness, a multiple neural network (MNN) model is presented to develop dynamic soft sensors. The a...Since in practical industrial applications, soft sensors based on static models often lead to low accuracy and poor robustness, a multiple neural network (MNN) model is presented to develop dynamic soft sensors. The approach is applied to building a dynamic soft sensor estimating the production component in a simulated binary distillation unit and good results are obtained.展开更多
现有的基于特高频（UHF）信号的局部放电时延定位方法研究重点多集中于提高时差的计算精度,而对系统定位误差的校正鲜有涉及。为此,论文直接从系统定位误差入手,提出了一种基于多神经网络的定位误差修正算法。在极径r∈[2 m, 6 m]、r∈[...现有的基于特高频（UHF）信号的局部放电时延定位方法研究重点多集中于提高时差的计算精度,而对系统定位误差的校正鲜有涉及。为此,论文直接从系统定位误差入手,提出了一种基于多神经网络的定位误差修正算法。在极径r∈[2 m, 6 m]、r∈[6 m, 12 m]及r∈[12 m, 18 m]这3个区间分段内建立了相应的误差补偿网络,利用有限个标定点的时延误差来训练径向基（RBF）神经网络,以模拟系统定位误差的分布特性,并对实际定位结果进行修正。仿真及实验结果表明,通过误差补偿网络的修正,提高了定位精确度、降低了定位结果的离散程度,最终可将定位距离误差控制在0.5 m以内,方向角误差控制在6°以内。研究结果验证了所提算法的误差修正能力。展开更多
文摘Since in practical industrial applications, soft sensors based on static models often lead to low accuracy and poor robustness, a multiple neural network (MNN) model is presented to develop dynamic soft sensors. The approach is applied to building a dynamic soft sensor estimating the production component in a simulated binary distillation unit and good results are obtained.
文摘现有的基于特高频（UHF）信号的局部放电时延定位方法研究重点多集中于提高时差的计算精度,而对系统定位误差的校正鲜有涉及。为此,论文直接从系统定位误差入手,提出了一种基于多神经网络的定位误差修正算法。在极径r∈[2 m, 6 m]、r∈[6 m, 12 m]及r∈[12 m, 18 m]这3个区间分段内建立了相应的误差补偿网络,利用有限个标定点的时延误差来训练径向基（RBF）神经网络,以模拟系统定位误差的分布特性,并对实际定位结果进行修正。仿真及实验结果表明,通过误差补偿网络的修正,提高了定位精确度、降低了定位结果的离散程度,最终可将定位距离误差控制在0.5 m以内,方向角误差控制在6°以内。研究结果验证了所提算法的误差修正能力。
