Performance evaluation of hybrid ANN based time series prediction on embedded processor

Complex embedded systems application exhibit time-varying workload which requires continuous resource adaptivity. Workload prediction has been successfully achieved through a hybrid model of NARX Recurrent Neural Networks combined with Self Organizing Map (SOM). This paper presents the performance evaluation of this hybrid time series prediction on embedded processors as an alternative to dedicated hardware. Achieved results demonstrate the potential of this approach for heavy workloads such as parallel applications. This solution is prone to extension to MPSOC.