058 A system for controlling a plant
The enumerated difficulties of realising a machine system of control can be completely overcome, provided we do not try to build an ideal system capable in any situation of finding instantaneously and implementing a strictly optimal control. In reality we do not require a system to be optimal but to be satisfactory, which will at least perform a control no worse than man would. The potentialities of man as regards receiving and processing information are limited, and even the most talented leader takes decisions on the basis of a relatively small number of factors and very rough forecasts, and spends a relatively long time in making decisions. This means that he cannot obtain a strictly optimal control. Nevertheless, plants and other complex organised systems controlled by man exist, and many of them develop successfully. This means that machine control systems may also prove viable if they perform the control functions, if not ideally, at least no worse than human beings do.
One possible structure of a machine system for controlling a plant is given in Fig. 17.3. This system is built up of learning, homomorphous (simplified) models of plants and media of various orders. The simplest and most quickly changing models of the first rank are intended for processing relatively small flows of information and to provide, quickly, control signals calculated to operate over short time intervals. The models of the second rank take into consideration more data, and calculate the control actions over a longer period of time. The models of the higher ranks work still more slowly but produce control actions for even longer periods. The models of the plant and of the medium of all ranks interact and generate the control signals of the group A, intended for keeping the plant in order, and the group B, for accomplishing its functions with respect to the medium.
In order that models change in accordance with changes in the properties and characteristics of the plant and the medium, they must be made adaptive, and for this purpose principles can be used which were described in Chapter 11. The ‚initiative‘ of a machine system, exemplified by a sufficiently large variety of possible solutions, can be achieved by introducing a moderate number of random changes from a generator of random signals.
The same principle can be applied for controlling not only the production process of a plant but also other complicated artificially organised systems which must function in a stable way under changing conditions.
Such systems may operate autonomously over a certain period of time which can be longer, the greater the number of ranks in the control hierarchy. However, from time to time the operation of the system has to be corrected by people who are responsible for ensuring that the system performs in a manner useful to society.