Geoscience Reference
In-Depth Information
In this study, MLP is designed in three layers including an input layer, two hid-
den layers and one output layer. In the input layer we have 16 neurons and in the first
and second hidden layers we have used 10 and 8 neurons respectively. In the output
layer we have 5 neurons representing each risk level. We have used backpropagation
algorithm as the training method with adjusted training parameters (momentum and
learning rate) and used sigmoid activation function in all layers of MLP.
In the test phase we have used 1.000 randomly selected records from universal
set which are not used in training phase to show the success rate of MLP accu-
rately. The best results have been obtained when the learning rate is set to 0.1 and
momentum is set to 0.5. The success rate shows the percentage of correctly pre-
dicted records. In the study, best prediction result has been found as 93.8 %.
4.3 3D Simulation of Intelligent Indoor Individual
Evacuation Model
The intelligent routing engine which works integrated with our proposed MLP
model is the most important part of our intelligent indoor evacuation model and is
responsible to produce real time instructions for the users to assist them accurately
till they arrive destination. For testing the evacuation model we have used our 3D
GIS based implementation presented in our previous study (Atila et al.
2012
). We
have done the required coding to integrate our proposed MLP with routing engine
for evacuation purpose. Our proposed simulation environment produces the input
variables (e.g. temperatures, fire growth rates, populations, visibilities etc.) needed
by MLP model for each link in the transportation network within a scenario.
In the beginning the acceptable risk level of the system is set to be “LOW”.
It means that, with the start of the evacuation, only links having “VERY LOW”
and “LOW” risk levels will be able to use to find a shortest path to the destination
and the others will be avoided. The flow chart given below explains the intelligent
evacuation process briefly (Fig.
13
).
4.4 Fire Evacuation Scenario
The scenario has been constructed in 3D model of Corporation Complex in
Putrajaya, Malaysia. We are able to run our simulation within predefined scenarios
stored in database. In the simulation, colors assigned on the links of network have
different meanings. Yellow color indicates a burning fire on a link but the link can
still be used. Gray color indicates smoke existence but the link can still be used.
While red colors indicates that the associated link is not in use any more, black
color shows the shortest path found by the system.
So, let's see how our intelligent evacuation model reacts in case of fire acci-
dent. Assume that a user has been in the 8th floor of a 10 floor building when the
