A QUEUEING NETWORK MODELLING APPROACH TO THE ANALYSIS OF HUMAN ATTENTIONAL PROCESSES.

摘要

In this thesis, a mathematical model for the allocation of attention is developed and an experiment which allows the testing of analytical predictions of the model is presented.; The motivation for mathematically modelling human behaviour stems from the need to quantify as much as possible human interventions in complex man-machine systems in order to obtain some global evaluation of the efficacy of the systems. Monitoring numerous indicators of the state of an industrial process, a nuclear power generator for example, is typical of such human intervention and the allocation of attention in a monitoring activity is the more specific behaviour modelled in this work, although the model developed could be extended to other attentional situations.; Regarding the various indicators under a human monitor's supervision as competing for his attention suggests the use of queueing theory for modelling the allocation of attention: the indicators constitute the "customers" who receive service as they are observed. More specifically, a queueing network with different types of customers is considered, with the possibility for a customer to change classes after each service. This last feature reflects the possibility of change in the urgency of observation of an indicator, depending on the particular evolution of that indicator. Stationary probabilities for the state of the network are derived and used to obtain the distribution of the waiting time of each instrument before it can be observed again. This waiting time is important because of the eventuality of a system failure escaping the monitor's attention.; An experiment of continuous monitoring of a panel of eight dials presenting random signals is described together with the recording techniques used to collect data on the behaviour of the monitor as related to the evolution of the system. The sequence of the monitor's eye fixations constitutes the main source of information on his behaviour and is recorded with the help of electro-oculography. Two subjects took part in the experiment, the goal of the modelling being the characterization of individual behaviour.; The data collected during the experiment are used to test statistically the assumptions and predictions of the model. Although some assumptions are not rigorously satisfied, supportive results are obtained. Simulation estimates moreover confirm these results. Possible extensions of the model are also discussed.