Title: Analysis, supervision, and control of a network of biological/chemical data paths using multi-step forecasting
Abstract:
It is assumed that there is given a network of biological/chemical paths. The network consists of nodes that contain historical data. Regression models are established between all nodes in the network. They are connected in each path. The analysis can start anywhere in the network at one or more nodes. If the data matrix, where one starts, is denoted by X, forecasts are developed for all nodes in all the paths that follow from X. Significance tests are used to determine how far reliable forecasts (from X) can be established along a given path. Methods of linear regression analysis are extended to regressions in data paths. Graphic methods, well-known in PLS Regression, are used to analyze the dependence between data blocks in a path. They illustrate how far and how well we can carry out forecasts. The forecasts show us what may be expected for later data blocks. This analysis can be carried out for all paths that follow from X. This can be used to supervise the biological/chemical processes. In practice, more than one model can be running. Control screens can be used to display the forecasts so that the engineers can see how the processes are performing in different parts of the network. This makes it possible for instance in a chemical production plant to model all processes. The forecasts show what may be expected. The results can be compared with the historical data and they can be used to evaluate the quality and performance of the processes. It is shown how a control center can be established that contains both the supervision of all processes and control of individual processes.