CAPÍTULO 3. Diseño y simulación
3.4 Simulación del diseño de red
3.4.4 Análisis de los resultados de la simulación
Table 6.1: Legend for model component symbols used in the SIMUL8 software package. The four primary model components I used for my model are given above in Table 6.1. Alongside these components, Visual Logic was used to make the model more flexible and enable dynamic behaviour. Full details of the visual logic I included has been given in Appendix C.
Figure 6.2: Layout of the SIMUL8 model. The thicker routing lines do not represent any difference to thinner lines. The consistent nodes for both the NGOs and the Partnership represent a consistent increase in their influence over time (using an exponential distribution
Figure 6.2 shows my current model, constructed in SIMUL8, for the public response to siting a GDF in Cumbria. This model follows my initial model plans quite closely, but there are significant portions of the model that are contained within the visual logic, which can be found in Appendix C. In particular, the current activity for the NGOs, Partnership and government incidents have very strong impacts on the probability distribution used for routing from the decision points. Each of these are used to calculate the current weights of the opinion states for each community. These weights are then used to calculate the routing probabilities from the decision points, taking into account the label values of the work item (multipliers are applied to the weights according to the label value). Also keep in mind that every work item in the Opinion States represents 100 like-minded individuals.
6.3.2 Model Construction
While this section contains information on how the model was constructed, I do not give specific details of every element of the model. However, Appendix C contains a detailed description of every important node included in the model.
6.3.2.1 Opinion Weight Variables
To introduce feedback into my model while taking into account the current events of the system, I decided to use base opinion weights for each of the opinion state. These base opinion weights are to some extent like the base flow rates that were introduced for my system dynamics model. Each community has an opinion weight for each of the five opin- ions, totalling 15 weights for all three communities. These reflect how visible the overall support for each opinion within the community is to the rest of the community according to the current events. For example, when the NGOs are more active than the partnership, the negative opinion weights receive slightly higher ‘event scaling’ than the positive opinion weights. Each of the opinion weights would also be scaled by the current government trust.
These opinion weight variables were updated every week and only change at the end of the week. This means that the opinion weights were fixed for each week according to the events active in the model at the start of that week. Ideally, I would update this whenever an individual arrives at the decision point to change their opinion but this would be far too costly for the simulation. The scaling factors I used here were based offthe same literature my system dynamics model used, and calibration according to the survey results from the MRWS Partnership. For example, the ‘Strong’ opinion weights were defined the same as the positive or negative opinion weights, with an additional scaling constant to represent less people holding these stronger opinions in the surveys. The DES model was designed in
this way, as it was a more natural way to model the scenario in DES, from the perspective of a DES modeller, rather than the opinion movements of my SD model.
6.3.2.2 Time-check Logic
As mentioned before in Section 6.2.6, the opinion weights were updated every 7 days and that update is done through the time-check visual logic in SIMUL8. This repeated a set of commands every 7 days. The details of the opinion weight updating, constant values, and other updating rules can be found in Appendix C. Alongside the opinion weights, other vari- ables were also updated on this time cycle. For example, the measurement of government trust was slightly altered from its previous value, and the relative NGO and Partnership ac- tivity were stored. Results were updated in an in-built spreadsheet, and finally the service time distributions were updated for each of the opinion states. As mentioned previously in the chapter, these service time distributions were all exponential, and so the parameter of this distribution changed on each time-check. Each opinion state had a defined constant for the exponential parameter, which may be scaled by the current NGO and Partnership activity. These constants were derived from the literature and calibrations of the model to promote realistic behaviour.
In particular, if the NGO and Partnership were inactive, the parameters remain at the con- stant level. However, if there was a large amount of NGO and Partnership activity, then the service time distribution parameters were reduced according to the amount of activity. This was done to promote more uncertainty for individuals as the process continues (as the NGOs and Partnership are becoming more active), making individuals visit the decision point more often later in the process. This behaviour was supported by the MRWS Partner- ship survey results, which suggested individual activity increases as the process progressed. The precise definition of the service time distribution can be found in Appendix C, although the amount of time between arrivals tended to be, on average, between 1 change every 3 months (late in the process with high engagement) to 1 change every 1.5 years (early in the process with low engagement).
6.3.2.3 Decision Node
The decision node was where much of the model complexity was contained. When a work item arrived at the decision node, they must make a decision on whether to change their opinion. This decision was made according to a probability distribution that is derived from both the current labels of the work item, and the current opinion weights of that work item’s community. This adjustment to the current opinion weights were made according the work item’s labels, so that each probability was better characterised to each work item’s current
circumstance (e.g. knowledge level, last opinion). In summary, when a work item decides to change opinion, the current base opinion weights are noted, and then adjusted according to the work item’s situation to create the probability distribution used to select the next opinion state.
The work item’s labels adjusted the base opinion weights as follows: the community label was used to select the correct base opinion weights for the work item’s community. The knowledge label was used to scale up the positive base opinion weights by 25%, if the work item was knowledgeable, else it increased the negative base opinion weights by 25% (this change is supported by the MRWS survey data). The last opinion label was used to increase the chance that the work item stays in the same opinion state they were last in. As time progresses, the chance that a work item remains in the same state decreases. This was to represent that individuals would get more involved with the process as time went on.