Basic Case Results
In Figure 9.1, there is an example of the first 10 consecutive choices that the fourth agent made in order to travel for a leisure activity, when it was weekend and nonrush hour. Both the overall expected utility of the chosen activity profiles and the conditional aspiration values of that activity type, under this context, are depicted in the graph. Specifically, after four unsuccessful exploration efforts, the agent lowers its conditional aspiration value, so that it becomes more realistic. Then the exploitation option becomes satisfactory enough to be selected on simulation day
Figure 9.1: A characteristic example of ten consecutive choices of the fourth agent for leisure activity, when it is weekend and non-rush hour.
Figure 9.2: Fluctuation of aspiration values of the second agent over time.
t = 63 and to constitute the habitual option of that agent for this activity type and under this context.
In Figure 9.2, the aspiration values, that the second agent holds, are depicted. Specifically, the conditional aspiration values for work, shopping and leisure activity type, as well as the aspiration for the long-term decisions and the actually experienced utility of the long-term decisions, are included. Initially, those aspiration values are adjusted according to the experiences of the agent and after some point of time, they are stabilised. The deviation between the aspiration for longterm decisions and the experienced utility of the long-term decisions represents stress at the long-term time horizon and if it exceeds σ2 value, the agent becomes 4 awake'.
Effect of Memory-Activation Parameters
Effect of λ2 parameter
Parameter λ2 is the retention rate, included in the forgetting part of the activation level update equation (Eq. (9.13)). As λ2 decreases, memory activation fades out very fast and therefore the agents 'stick' less to their previous experiences. Thus, while λ2 decreases, the agents develop habits with bigger difficulty, resulting in a decreasing frequency of habitual behaviour (Figure 9.3a). However, as λ2 decreases, more exploitation choices take place (Figures 9.3a and b). This is because memory activation decreases very fast and the agents need to exploit better their choice-set, as they don't have a strong habit indicator. On the other hand, as λ2 increases, the agents develop strong habits and do not exploit the rest of their choice-set enough. This implies that even if there were better options than their habits, they are discarded from memory very fast. However, if at some point of time these strong habits are not satisfactory any more, the choice set hardly contains any other alternatives and the agents have to explore the area in order to find a better solution. Thus, when parameter λ2 increases, exploratory behaviour increases, as well (Figure 9.3b). Finally, choice-set size is also affected, as it tends to increase, due to the increasing number of explorations.
Figure 9.3a and b: Effect of λ2 parameter on frequency of choice modes (habit, exploitation, exploration) and lowering aspiration incidents.
As λ2 decreases, the agents cannot develop easily habits and they tend to better exploit their choice set. This results in a higher mean overall expected utility of the habitual and exploitation choice modes (Figure 9.4). Consequently, the overall expected utility of their choice set also improves (Figure 9.5). Specifically, when parameter λ2 decreases, the agents better exploit their choice-set and the alternatives that perform well have a lower chance to be discarded from memory. Finally, aspiration values are also affected. Specifically, they also tend to increase, when parameter λ2 decreases, because of the increased overall expected utilities of the habit and exploitation choice modes.
Figure 9.4: Effect of λ2 parameter on overall expected utility of habit, exploitation and exploration choice mode.
Figure 9.5: Effect of λ2 parameter on choice set overall expected utility.
Thus, when the retention rate of memory activation is too high, habitual behaviour becomes too strong and the agents do not manage to exploit well their choice set. It also appears that when activity-travel behaviour is driven by strong habits, there is a decline of welfare over time and an increase of exploration effort.