Amnesic Multi-sensor Fusion Algorithm for Streaming Time Series

Since more and more real time multi-sensor system generated continuous measurement values, a novel amnesic fusion algorithm is proposed. In the case of any prior knowledge is unknown, the algorithm takes use of the exponential function to build the support degree matrix, so each sensor´s weight coefficient of different time can be obtained. Based on the fact that in many domains recent information is more useful than older information, the user-specified piecewise constant amnesic function is defined and ensured the correctness of weight coefficient re-assignment and the reduction of computation and storage complexity. Three thermocouples with observation noise distribution unknown are used to detect the temperature of a constant temperature box to verify the effectiveness of this algorithm.