Abstract
The control of a modem power generating station is a complex task involving the acquisition and processing of a large amount of data. This involves the processing of data from transducers or other inputs which then produce the desired outputs for actuators and displays etc. The advent of small cheap digital data processing systems has made it economically desirable and indeed feasible to implement distributed computer control schemes. The overall control of the station can be achieved by an interconnected set of such computer systems, each computer being at the node of a communications network. The actual control functions are implemented as a number of co-operating modular programs resident in each of the control nodes. It is assumed that the computers will be functionally similar ( hardware and software ) and that the characteristics of each module (task) such as the CPU loading and the inter-task communication requirements are known a priori. This work investigates the assignment of these tasks such that the distributed computer network uses the minimum number of computers and that the overall inter-computer communication is minimised. However, this overall objective is influenced by a number of technical and operational constraints which are used to formulate a series of mathematical models that progressively include more aspects of the problem. The application of various linear and non-linear optimisation techniques to the solution of these models is investigated. Three independant methods of optimisation are investigated to solve the computer control network problem. In each case, the aim is to construct a simple model based on certain aspects of the problem and then extend the model to include all other aspects. A complete mathematical model which applies the standard methods of optimisation is presented. It is claimed that these formulations are original. It is shown that the complete network design problem is difficult to solve efficiently using standard methods of optimisation, because of the size and the complexity of a practical problem. Hence, the central component of this research has been the development of an algorithm to solve practical network design problems. This algorithm is claimed to be original and is computationally more efficient than the standard methods of optimisation for this type of problems. The basic steps of this algorithm are to decompose the problem and then interactively solve the less complex subproblems. It is shown that this algorithm used interactively will provide a feasible solution. This work contributes to the research into the design of distributed computing systems for process control applications, undertaken by the Central Electricity Research Laboratory (C.E.R.L., Leatherhead, Surrey). In particular it makes a major contribution to the objective of producing a design aid for such computing systems.