Analysis and Identification of Mineral Process Plants by Use of Singular Spectrum Analysis and Neural Networks

"Owing to environmental constraints, safety considerations and competitive pressures, the development of model-based process control and other automated systems have experienced considerable growth in the last few decades. Reliable process models are usually critical to the successful implementation of these systems. In complex processes, these models are often constructed from process data, as modelling from first principles may not be feasible or cost-effective.Even so, modelling or identification of nonlinear systems from historic plant data can be a daunting task and in the process industries the problem is complicated by the presence of noise from various sources, nonstationarity of the data, intermittence, such as observed in particulate flows, etc. Traditional methods, such as frequency analysis and linear modelling do not handle these systems well. Similar problems arise with the application of nonlinear theory developed over the last few decades, where much of the analysis depends on embedding of the data in a phase or pseudophase space, since these methods were not originally designed to deal with noisy systems.With singular spectrum analysis, the data are embedded in a high-dimensional reconstruction, followed by the introduction of a new coordinate system, where the origin is moved to the centroid of the reconstructed system states and the axes are represented by the dominant principal components of the states (points)."