On nonparametric algorithms of object groups control

The paper is devoted to the identification and control of a group of both identical and varied objects. First the con- trol system of a local object is considered. Further on, all random factors acting in the measurement channels of an object are omitted and only the relationships between objects are preserved. Nonparametric algorithms of identification and control are given in a rather general form. The main attention is paid to different relationships between local objects. First the simplest case is considered, that is, a serial technological chain for whose control it is offered to use an external circuit of control when control influences in an external circuit play the role of a setting influence of local systems. The group of objects of a parallel-serial type is the following in complexity and here some feature inherent in many variables figuring in the corresponding section of a group of objects shows itself. This essential feature is that the output variables are measured with different discreteness, and the discreteness of control can significantly vary, includ- ing hours, shifts, days and weeks, depending on the nature of the production process and the scheme of measurement and control of corresponding variables. For example, the fineness of grinding in the production of cement is measured several times per shift, and the activity of cement showing its brand is determined only after 28 days. This circumstance should be taken into account when developing computer systems for identification, control and decision-making. The most difficult are such groups that fit into the scheme of interrelated objects, where feedback takes place in the technological process. The external control circuit can have a hierarchical structure, where the number of hierarchies is related to technical-economic, production-economic factors and in active systems to social factors as well. Control of such complexes is a complicated task, where the application of the theory of adaptive and learning systems is justified. This paper considers a class of algorithms of a non-parametric type, i.e. algorithms oriented to the case when informa- tion for describing objects is insufficient for parameterization, i.e. selection up to a set of vector parameters. Similar algorithms are presented in a rather general way. As a result, some cases of using control systems for groups of objects for thermal power plants are considered, and condensed information is provided for the metallurgical processing of the Norilsk Combine.