【摘要】 对基于反馈神经网络的化工过程动态数据校正方法进行了研究,实现了自反馈增益的网络结构和动态反向传播算法。连续搅拌槽反应器实例的应用结果表明,反馈网络能快速、有效地校正动态过程测量数据。该方法仅依赖于历史测量数据,无需掌握过程本身的精确模型及测量噪声的参数,克服了应用过程机理模型对化工动态过程进行数据校正的局限性。更多还原

【Abstract】 Chemical processes are highly non-linear systems exhibiting complex time-dependent behavior.Modeling of these dynamics using first principle is not always possible,which brings difficulties to process data rectification.An alternative dynamic data rectification method based on recurrent neural networks(RNN)was studied in detail.We demonstrated the application of modified RNN with self-feedback gains to rectify the simulated measurements obtained from dynamic continuous stirred tank reactor(CSTR).The results indicated that RNN could be trained to generate excellent dynamic process models and rectify dynamic measurements fast and efficiently.In contrast to conventional nonlinear programming and Kalman filter,RNN-based method was independent of rigorous mechanism models of process and information about noise parameters,relying only upon historical measurements.RNN offered a promising alternative method for dynamic process data rectification.更多还原