Sliding Window Based Feature Extraction and Traffic Clustering for Green Mobile Cyberphysical Systems

Algorithm 2

The traffic clustering algorithm.

Input: Feature vectors of all traffic distribution samples .

Output: Traffic patterns .

 (1) Normalize Feature vectors to .

  (2) Determine the number of traffic patterns by the average silhouette method.

  (3) Construct an affinity matrix with Gaussian kernel function, in which

    holds for and .

  (4) Define the diagonal degree matrix . Normalize the affinity to , and .

  (5) Compute the first eigenvectors of . Construct a matrix .

  (6) Construct a matrix from by normalizing the rows of to norm 1, and .

  (7) Treating each row of as a point, cluster them into traffic patterns by -means.

  (8) Assign the original feature vector of traffic distribution sample to traffic pattern according

   to the assigned label of the row of the matrix .

  (9) Compute the features of traffic patterns, and .