/// <summary>
/// 概率矩阵其实就是优选取计数器
/// </summary>
private void RefreshProbabilityMatrix(ProbabilityMatrix probablityMatrix, List <OptimizedResult> resultList)
{
foreach (OptimizedResult res in resultList)
{
for (int i = 0; i < res.Tasks.Count; i++)
{
probablityMatrix[i, res.Selections[i].Item1]++;
probablityMatrix[i, res.Selections[i].Item1, res.Selections[i].Item2]++;
}
}
}
/// <summary>
/// 随机生成一个任务序列
/// </summary>
private Tuple <int, int>[] GenerateSequence(IList <IList <IList <Task> > > taskMatrix, Task[] tasksArray,
Random random, ProbabilityMatrix probabilityMatrix,
double selectDiffElement)
{
Tuple <int, int>[] selectionArray = new Tuple <int, int> [tasksArray.Length];
//Tuple->kind index->source index->count
int[][] indexSelected = new int[taskMatrix.Count][];
for (int i = 0; i < taskMatrix.Count; i++)
{
indexSelected[i] = new int[taskMatrix[i].Count];
for (int j = 0; j < taskMatrix[i].Count; j++)
{
indexSelected[i][j] = taskMatrix[i][j].Count;
}
}
int index = 0;
List <int> probIndex = new List <int>();
List <double> probList = new List <double>();
while (index < tasksArray.Length)
{
double sum = 0;
probList.Clear();
probIndex.Clear();
//加入择异因子
for (int i = 0; i < taskMatrix.Count; i++)
{
bool lastSameHint = false;
if (index > 0 && selectionArray[index - 1].Item1 == i)
{
lastSameHint = true;
}
if (indexSelected[i].Sum() > 0)
{
double proValue = lastSameHint ? (probabilityMatrix[index, i] / selectDiffElement) : probabilityMatrix[index, i];
sum += proValue;
probIndex.Add(i);
probList.Add(proValue);
}
}
int selectKind = RandomSelect(random, probIndex, probList, sum);
probList.Clear();
probIndex.Clear();
sum = 0;
for (int i = 0; i < taskMatrix[selectKind].Count; i++)
{
if (indexSelected[selectKind][i] > 0)
{
sum += probabilityMatrix[index, selectKind, i];
probIndex.Add(i);
probList.Add(probabilityMatrix[index, selectKind, i]);
}
}
int selectSource = RandomSelect(random, probIndex, probList, sum);
int taskIndex = selectKind;
int sourceIndex = selectSource;
int itemIndex = taskMatrix[selectKind][selectSource].Count - indexSelected[selectKind][selectSource];
selectionArray[index] = new Tuple <int, int>(taskIndex, sourceIndex);
tasksArray[index++] = taskMatrix[taskIndex][sourceIndex][itemIndex];
indexSelected[selectKind][selectSource]--;
}
//设置后置复位任务关键节点
//for(int i = 0; i < tasksArray.Length; i++)
//{
// tasksArray[i].RecoverJoint = null;
// for(int j = i + 1; j < tasksArray.Length; j++)
// {
// if(selectionArray[i].Item1 == selectionArray[j].Item1)
// {
// tasksArray[i].RecoverJoint = tasksArray[j].GetUsingRailJoint();
// break;
// }
// }
//}
return(selectionArray);
}
/// <summary>
/// 启发式方式
/// </summary>
private OptimizedResult HeuristicMethodOptimize1(List <Task> tasks,
int batchSize, int refreshSize,
int maxIterationCount, double errorRatio,
double eMatrixRatio, double selectDiffElement)
{
OptimizedResult result = new OptimizedResult();
result.TimeCost = Double.MaxValue;
IList <IList <IList <Task> > > taskMatrix = ParseTasks(tasks);
Task[] tasksArray = new Task[tasks.Count];
Random random = new Random();
OptimizedResult[] batchResult = new OptimizedResult[(int)(batchSize * (1 - eMatrixRatio))];
OptimizedResult[] eBatchResult = new OptimizedResult[(int)(batchSize * eMatrixRatio)];
List <OptimizedResult> sortedList = new List <OptimizedResult>(batchSize);
ProbabilityMatrix dProbablityMatrix = new ProbabilityMatrix(taskMatrix, tasks.Count);
ProbabilityMatrix eProbablityMatrix = new ProbabilityMatrix(taskMatrix, tasks.Count);
String output = "{0}::当前最优:{1:0.00},平权--均值:{2:0.00}, 标准差:{3:0.00}。动态--均值:{4:0.00}, 标准差:{5:0.00}。";
Debugger debugger = new Debugger();
int batchIndex = 0;
int iteration = 0;
while (maxIterationCount != -1 && iteration < maxIterationCount)
{
//计算一个，先搞平权的batch数量
for (batchIndex = 0; batchIndex < eBatchResult.Length; batchIndex++)
{
Tuple <int, int>[] selectArray = GenerateSequence(taskMatrix, tasksArray, random, eProbablityMatrix, selectDiffElement);
double timeCost = CalculateSequence(tasksArray.ToList());
iteration++;
OptimizedResult res = new OptimizedResult();
res.TimeCost = timeCost;
res.Tasks = tasksArray.ToList();
res.Selections = selectArray;
eBatchResult[batchIndex] = res;
if (timeCost < result.TimeCost)
{
result.Counter = 0;
result.Selections = res.Selections;
result.Tasks = res.Tasks;
result.TimeCost = res.TimeCost;
}
else if (timeCost == result.TimeCost)
{
result.Counter++;
}
}
for (batchIndex = 0; batchIndex < batchResult.Length; batchIndex++)
{
Tuple <int, int>[] selectArray = GenerateSequence(taskMatrix, tasksArray, random, dProbablityMatrix, selectDiffElement);
double timeCost = CalculateSequence(tasksArray.ToList());
iteration++;
OptimizedResult res = new OptimizedResult();
res.TimeCost = timeCost;
res.Tasks = tasksArray.ToList();
res.Selections = selectArray;
batchResult[batchIndex] = res;
if (timeCost < result.TimeCost)
{
result.Counter = 0;
result.Selections = res.Selections;
result.Tasks = res.Tasks;
result.TimeCost = res.TimeCost;
}
else if (timeCost == result.TimeCost)
{
result.Counter++;
}
}
//计算并记录各项指标
double avg = batchResult.Length == 0 ? 0 : batchResult.Average(b => b.TimeCost);
double s2r = batchResult.Length == 0 ? 0 : Math.Sqrt(batchResult.Sum(b => Math.Pow(b.TimeCost - result.TimeCost, 2)) / batchResult.Length);
double eAvg = eBatchResult.Length == 0 ? 0 : eBatchResult.Average(b => b.TimeCost);
double eS2r = eBatchResult.Length == 0 ? 0 : Math.Sqrt(eBatchResult.Sum(b => Math.Pow(b.TimeCost - result.TimeCost, 2)) / eBatchResult.Length);
debugger.StringAppend(iteration).StringAppend(" ")
.StringAppend(result.TimeCost).StringAppend(" ")
.StringAppend(eAvg).StringAppend(" ")
.StringAppend(eS2r).StringAppend(" ")
.StringAppend(avg).StringAppend(" ")
.StringAppend(s2r).StringAppend(" ")
.StringAppend("\n");
Debugger.WriteLine(String.Format(output, iteration, result.TimeCost, eAvg, eS2r, avg, s2r));
sortedList.Clear();
sortedList.AddRange(eBatchResult);
sortedList.AddRange(batchResult);
sortedList.Sort(new OptimizedResultComparator());
if (eMatrixRatio < 1)
{
//更新概率矩阵
RefreshProbabilityMatrix(dProbablityMatrix, sortedList.Take(refreshSize).ToList());
}
}
debugger.WriteTofile();
return(result);
}
