A Neural Network based Furnace Control System

In the future, control systems will become increasingly complex to meet more demanding requirements. Maintaining optimum performance may entail the precision control of dynamic systems whose model is unknown or highly uncertain, is highly complex (multidimensional and non-linear), and can vary with time. This will require a robust and adaptive control system, the capabilities of which are likely to be beyond those available with conventional techniques. Conventional data processing techniques are ideally suited to applications where the physics of the problem are understood or rules can be specified to describe the behaviour. Artificial neurl networks, in contast, are at their most powerful when applied to problems whose solution requires knowledge which is difficult to specify, but for which there is an abundance of examples. Most complex control problems are of this type in that examples of the system behaviour in response to different control system stimuli are redily availble. British Gas plc and SD-Scicon UK Limited have carried out an investigation of the application of neural networks to the model based control of an experimental furnace. For this, SD-Scicon developed a neural network model of the fumace using open-loop test data provided by British Gas, who incorporated the model into an explicit generalised predictive control scheme and carried out a performance evaluation. This paper describes the background to this work, describes the control system design including the neural network development and training, and presents and discusses the control system performance.