Multivariate Time Series Data Clustering Method Based on Dynamic Time Warping and Affinity Propagation

Algorithm 1

AP clustering (S)

Input: similarity matrix S between data points, the maximum iteration number cot, damping coefficient .

Output: cluster set .

S1: initialize the responsibility matrix r and the availability matrix to be zero matrix.

S2: Update the responsibility matrix .

S3: update the availability matrix .

S4: introduce the damping coefficient to reduce the possible vibration when updating information. Apply the weighted sum of times of the last iteration update value and times of the current information update value to assign each information. That is

S5: repeat steps S2 to S4. Stop the algorithm when the clustering results tend to be stable or achieve the preset iterative number cot.

S6: construct the matrix by adding the matrices and and further find the corresponding column where the value of each row is the maximum. is the cluster center of object .

S7: assign the objects with the same cluster center to the same cluster and obtain the finally cluster set .