工业过程指标的平滑交替辨识预报算法

A smooth alternate identification algorithm for industrial process index prediction

中文关键词  智能控制  复杂工业过程  运行指标预报  平滑交替辨识

英文关键词  intelligent control  complex industrial process  operational index prediction  smooth alternate identification

中文摘要

      针对复杂工业过程的指标预报问题, 本文提出一种基于数据的非线性系统平滑交替辨识算法. 交替辨识算法将系统的输入输出模型在工作点附近展开为线性模型和高阶非线性模型, 然后交替更新线性模型参数和非线性模型参数, 其中对于线性模型采用最小二乘辨识方法, 对于高阶非线性模型采用长短期记忆网络进行建模. 所提方法的创新之处在于, 对于实际系统中的噪声易导致线性部分辨识参数震荡的问题, 引入平滑因子来抑制震荡, 提高预测模型的稳定性能; 在非线性部分则引入压缩因子来调节在辨识过程中非线性部分的权重, 总体上提高了预报的准确性. 通过数值仿真验证了所提算法的性能, 并与其他方法进行了对比实验, 结果表明所提算法能够有效抑制辨识过程中的参数震荡, 并且取得更好的辨识精度.

英文摘要

      In this paper, an alternating identification algorithm with a smoothing factor for nonlinear systems based on data is proposed for the problem of index forecasting for complex industrial processes. The alternating identification algorithm expands the input and output model of the system into a linear model and a higher-order nonlinear model near the operating point. Then, the parameters of the linear model and nonlinear model are updated alternately. The least squares identification method is used for the linear model, and the long-short memory network is used for the higher-order nonlinear model. The innovation of the proposed method is that for the problem that the noise in the actual system is easy to cause the oscillation of the identification parameters of the linear part, the smoothing factor is introduced to suppress the oscillation. In the nonlinear part, the compression factor is introduced to adjust the weight of the nonlinear part in the identification process, which improves the accuracy of the forecast. The performance of the proposed algorithm was verified by numerical simulation and compared with other methods. The results show that the proposed algorithm can effectively suppress parameter oscillation in the identification process and achieve better identification accuracy.