Requirements Analysis
SP Bosbrand Specifications Team sp-bosbrand@cs.uu.nl
This specification is split up in four main thematic parts. In each of
the sections we present the list of expectations of the project provider and
also our proposed solutions to those problems. Technical details and aspects
of the implementation were not considered.
The world has certain global attributes.
- windpower
- winddirection
- humidity
These attributes are the same for all cells that constitute part of the
forest (`world').
Each static item has the following properties.
- fuel
- determines duration cell is on fire.
- threshold
- below which it burns
- state
- not burning, burning, burnt
- cost
-
The following static items will be implemented in the ``final'' version.
- Trees
- Tree, we only keep track of one kind of tree that has the
following attributes.
- Grass
- Grass.
- Water
- Our primary interest is that it doesn't burn.
- Road
- Promenade routes in the forest. These can be large enough so that
a bulldozer can drive on it, reducing the time of the travel. Unlike a
fire-line, fire can cross this cell.
- Fire-line
- We will use the word `fire-line' to describe the part of the
forest that has been cleaned by the bulldozers. Fire cannot cross this cell.
1
- Building
- This is expensive when burnt down so the controller should avoid
this (represents residential area, industrial zone, etc.).
- Depot
- The start point of the ground fire-fighting agents, deposit place
for bulldozers. They are expensive when burnt.
The coordinates and attributes of these cells are determined at runtime.
- Fire
- Represents a burning cell.
- Waterbomb
- Region that decreases the fire activity.
- Ground Agent
- Represents a bulldozer, which is able to create a
fireline, trough grass or trees. This agents starts in a depot.
- Air Agent
- Flying agent, drops waterbombs on the fire.
The behaviour of the agents is determined by the controller.
The simulation will be modeled using Cellular Automata (4 neighbours, no
cyclic boundary conditions). It determines the fireactivity for tree's and
grass using the cell's attribute(s). The propagation of the fire will be
stochastic.
The basic task for the controller is to direct all active agents.
It calculates subtargets for each agent and generates the most optimal route
for the agents. Evalution of subtargets is implemented using neural networks
(NN) of which the weights are `trained' using ESP.
The shortest route between the subgoals is determined using dynamic path
planning algoritmes (implemeted using ''A*'' or ''Dijkstra's kortste pad
algoritme'').
- Agents begin at selected location(s).
- The location of (sub)targets are dynamic and wil be relocated.
- When multiple agents are active, they will cooperate.
- An initial startpoint is chosen according to the four or eight
winddirections (N,O,Z,W).
- The fire-line is an closed polygon, the last subtarget coincides with
the first.
- The location of subtargets will be constantly re-evaluated.
- There will be only one fireline, even when using multiple agents.
There will be two kinds of agents, airborne units and ground units.
- Agents will not enter the fire.
- Subtargets are unique per agent.
- Agents are able to dig a fire-line, through grass or trees. Trees
require more time than grass.
- When possible the agents will use water as part of the fire-line.
- The fireline will be created as precisely as possible according to the
planned route.
- Units move faster on existing roads.
- Agents will drop waterbombs on strategic locations in the fire.
- Agents have to refuel water at an airport or in a lake.
- A waterbomb will only slow a fire down.
The path planner will plan the path from start location to first subtarget.
From there on it will plan from subtarget to subtarget. The path will be
optimized to minimalize costs. The demands are listed in ascending order,
specific to the costs of the actions.
- Riding on existing roads
- Riding on gras
- Riding trough trees
- Digging of grass
- Digging trough trees
The application clearly has two modes of operation, learning the
agent's NN using ESP and visualizing a single run after the
NN are trained. Therefor we will provide two seperate interfaces.
This interface provides statistical information on the population etc.
It also allows modification of the ESP parameters and related parameters
such as those of the random world generator.
It can be run autonomously and will store the (possibly intermediate) NN and
other results of the process for later examination.
The userinterface provides a convenient world editor for building customary
worlds.
It furthermore allows visualization and modification of parameters related to
the simulation (wind speed, direction, start point of fire, humidity).
This allows modifying (and saving/storing) configurations.
These can be used for training new NN (using the ``learning mode'')
or for demonstrating existing NN on.
Requirements Analysis
This document was generated using the
LaTeX2HTML translator Version 2K.1beta (1.48)
Copyright © 1993, 1994, 1995, 1996,
Nikos Drakos,
Computer Based Learning Unit, University of Leeds.
Copyright © 1997, 1998, 1999,
Ross Moore,
Mathematics Department, Macquarie University, Sydney.
The command line arguments were:
latex2html -dir /home/wires/Bushfire/website/html/docs/specs -split 0 ../../../tmp/docs/functionaldesign/specs.tex
The translation was initiated by Jelle on 2003-04-26
Footnotes
- ... cell.1
- This may be misleading because the same word is sometimes used to
refer to the concept of to the meeting point between the fire and the yet
untouched forest, these being the points where the fire has the highest
activity. We will refer to this last concept as `fire-front'.
Jelle
2003-04-26