Parallel Algorithm for Time Series Based Forecasting on OTIS-Mesh

Forecasting plays an important role in business, technology, climate and many others. As an example, effective forecasting can enable an organization to reduce lost sales, increase profits and more efficient production planning. In this paper, we present a parallel algorithm for short term forecasting based on a time series model called weighted moving average. Our algorithm is mapped on OTIS-mesh, a popular model of optoelectronic parallel computers. Given m data values and n window size, it requires ) 1 ( 5 n electronic moves + 4 OTIS moves using n2 processors. Scalability of the algorithm is also discussed.