On state estimation and neural networks in process engineering

Many approaches have been developed for estimating those variables which are difficult to measure online in industrial process situations. In this paper, two approaches that can be used to provide frequent and accurate estimates of process outputs which are subject to large measurement delays are outlined. The first is based upon linear adaptive techniques whilst the other makes use of a fixed parameter neural network model. The development and application of the two estimators is addressed. The results from recent industrial application studies and plant simulation studies serve to highlight the characteristics of the different philosophies taken in estimator design. Moreover, this allows a comparison of the performance capabilities of the two techniques