MANAGEMENT OF TIME-VARYING NORMAL STATES IN DYNAMIC SYSTEMS AND ITS APPLICATION

点击翻译按钮获取中文摘要

Management system for prediction and supervision of a state in a dynamic system using time series data obtained in plant operations has been discussed. The following functions were considered to be required to the system :Function (1) : Prediction of an uncertainly time-varying normal state in the future.Function (2) : Judgment of whether a present state is under normal operation of not.Function (3) : Setting of upper and lower limits of alarms according to operational conditions.Function (4) : Extraction of characteristics of abnormal states for fault diagnosis.From the experiments and simulations applying the proposed method to a tank-pipeline process, the following can be concluded that :1.Autoregressive exogenous model (ARX model) , which is a linear model, can be used for Function (1) , even if normal values of state variables in the weakly nonlinear process (tank-pipeline process) are uncertainly time-varying due to input fluctuations.2.3 types of neural networks (feedfoward, external reccurent, general reccurent) , which are nonlinear model, can be also used for Function (1). Then the general reccurent neural network gives the best prediction.3.Sequential probability ratio test (SPRT) based on the error residual between the measured value and the corresponding value predicated by Function (1) can be used for Function (2).4.Function (4) can be realized by the improvement of Function (3) so that 2 SPRT are simultaneously performed : one test examines whether the error residual is normal or higher than normal, the other examines whether it is normal or lower than normal.5.Function (3) remains as a future work, but may be feasible by the extension of Function (4).