public List <RubisStation> Calculate(List <OHCAEvent> eventList, int budget)
{
List <RubisStation> prevStationList = new List <RubisStation>();
List <RubisStation> nextStationList;
double epsilonTemp = 0.1;
double alpha = 0.995;
double bestSurvivalRate = 0.0;
int tempBudget;
int maxStations = (int)(budget / (Utils.STATION_PRICE + Utils.DRONE_PRICE));
//for (int stations = 1; stations <= maxStations; stations++)
int stations = 16;
tempBudget = budget;
// Step 1. Finds initial stations with a drone using K-Means
tempBudget = tempBudget - (stations * (Utils.STATION_PRICE + Utils.DRONE_PRICE));
KMeansResults <OHCAEvent> kMeansStations = KMeans.Cluster <OHCAEvent>(eventList.ToArray(), stations, Utils.KMEANS_ITERATION_COUNT);
prevStationList.Clear();
foreach (double[] d in kMeansStations.Means)
{
prevStationList.Add(new RubisStation(d[0], d[1], 2));
}
// Step 2. Assigns remaining drones to busy stations
/*int remainingDrones = (int)(tempBudget / Utils.DRONE_PRICE);
* while (remainingDrones > 0)
* {
* int mostBusyStationIndex = getIndexOfMostBusyStation(prevStationList);
* prevStationList[mostBusyStationIndex].droneList.Add(new Drone(prevStationList[mostBusyStationIndex].stationID));
* remainingDrones--;
* }
*/
// Step 4. Simulated Annealing
double currentTemp = 100.0;
int iteration = 0;
double prevSurvivalRate = GetOverallSurvivalRate(ref prevStationList);
double delta = 0.0;
double nextSurvivalRate = 0.0;
while (currentTemp > epsilonTemp)
{
iteration++;
// Near search using local optimization
nextStationList = MoveOneStepToBestDirection(prevStationList, prevSurvivalRate);
nextSurvivalRate = GetOverallSurvivalRate(ref nextStationList);
delta = nextSurvivalRate - prevSurvivalRate;
if (delta > 0)
{
CloneList(nextStationList, prevStationList);
prevSurvivalRate = nextSurvivalRate;
}
else
{
// Even if worst, choose it randomly according to the current temperature
double probility = new Random().NextDouble();
if (probility < Math.Exp(-delta / currentTemp))
{
// Far search using random placement
nextStationList = FindRandomStationPlacement(prevStationList, 0);
CloneList(nextStationList, prevStationList);
prevSurvivalRate = GetOverallSurvivalRate(ref prevStationList);
if (prevSurvivalRate < bestSurvivalRate * 0.9)
//double r = new Random().NextDouble();
//if (r < 0.1)
{
/*
* kMeansStations = KMeans.Cluster<OHCAEvent>(eventList.ToArray(), stations, new Random().Next(50, 100));
* foreach (double[] d in kMeansStations.Means)
* {
* prevStationList.Add(new RubisStation(d[0], d[1], 2));
* }*/
Random rand = new Random();
bool okay = false;
while (!okay)
{
int pos = rand.Next(0, 990000);
kMeansStations = KMeans.Cluster <OHCAEvent>(simulator.GetSimulatedEvents().GetRange(pos, 10000).ToArray(), stations, Utils.KMEANS_ITERATION_COUNT);
okay = true;
foreach (double[] d in kMeansStations.Means)
{
if (d[0] == 0.0 || d[1] == 0.0)
{
okay = false;
break;
}
}
}
prevStationList.Clear();
foreach (double[] d in kMeansStations.Means)
{
prevStationList.Add(new RubisStation(d[0], d[1], 2));
}
/*
* remainingDrones = (int)(tempBudget / Utils.DRONE_PRICE);
* while (remainingDrones > 0)
* {
* int mostBusyStationIndex = getIndexOfMostBusyStation(prevStationList);
* prevStationList[mostBusyStationIndex].droneList.Add(new Drone(prevStationList[mostBusyStationIndex].stationID));
* remainingDrones--;
* }
*/
prevSurvivalRate = GetOverallSurvivalRate(ref prevStationList);
}
}
}
// Keep the best solution
if (prevSurvivalRate > bestSurvivalRate)
{
CloneList(prevStationList, stationList);
bestSurvivalRate = prevSurvivalRate;
}
// Cool-down
currentTemp *= alpha;
Console.WriteLine("[" + iteration + "] Stations = " + stations + ", Temp.: " + String.Format("{0:0.000000000000}", currentTemp) + "℃ " +
"Best = " + String.Format("{0:0.000000}", (bestSurvivalRate * 100)) + "% " +
"Current = " + String.Format("{0:0.000000}", (prevSurvivalRate * 100)) + "%");
}
return(stationList);
}