A Study of Cognitive Skills in Complex Systems

Abstract

The work described in this thesis is concerned with investigating certain aspects of human behaviour in control tasks requiring cognitive skill. The first chapter reviews evidence in the process control literature on, (i) the characteristics of control behaviour, (ii) development of process control skills, (iii) individual differences, (iv) factors that affect performance, (v) structure of operators' behaviour, (vi) models of the process operator, (vii) methodology. The next chapter describes a preliminary field study of the behaviour of gas grid control engineers. The investigation was carried out using observations and reports from the controllers as the main source of data. The results indicated that the verbal protocol technique might give useful and more detailed data on control behaviour. Chapter 3 investigates the controllers' behaviour using verbal protocols. Previous methods of analysing protocol data are reviewed. Three methods are used to analyse the protocols of the grid controllers. The results indicate that: (a) peak activity occurred during the afternoon shifts, (b) individual differences between controllers were associated with different types of predictive behaviour, (c) some of the controllers appeared to make more accurate predictions than others. The next chapter describes a comparative study of electricity grid control engineers. The procedures for investigation are similar to those of the previous two chapters. Results from the three types of analysis indicate that, (a) morning was the most active part of the day, (b) only 32% of goal orientated behaviour was concerned with predicting and decision making compared with 88% in the gas task. A detailed analysis of decisions shows several types of decision behaviour, which are combined into a general model of decision behaviour. Chapter 5 describes an experiment in which a simulation of a simplified version of the gas control task is used to examine the effect of certain types of predictive information on the performance of naïve subjects, and to test their understanding of particular aspects of the task. Results show that although subjects’ performance improved over trials, it was not affected by the additional predictive information. Results also show that only knowledge about certain task variables seemed to be relevant to good control performance. Chapter 6 describes another experiment using the simulator in which naive subjects are given predictive information but for more trials than in the previous chapter. Changes in behaviour due to skill acquisition during the trials are investigated using verbal protocols. The results indicate that even with twice as many trials subjects could not learn to use the predictive information effectively. The protocol data shows an increase in purposive behaviour and in particular an increase in predictive behaviour. Chapter 7 describes an experiment in which the grid controllers used a simulated version of the gas control task. The purpose of the experiment was to investigate controller behaviour under more difficult and controlled conditions, using verbal protocols. The results show that behaviour on the simulator was similar to behaviour on the real task with slight modifications to cope with the more difficult task. Performance was assessed using several criteria; only small differences were found between the controllers. Chapter 8 summarises and discusses the results of the experiments and field studies and relates these to the various areas of research reviewed in Chapter 1. In particular the possibility of fundamental cognitive abilities underlying control behaviour is discussed, as well as possible limitations in the information combining processes of the prediction and decision procedures. A suggested approach for further work on cognitive skills is also mentioned.