Network-Based Layered Architecture for Long-Term Prediction

In time series forecasting, a challenging and important task is to realize long-term prediction. This paper proposes a layered architecture based on backpropagation neural network. The proposed layered architecture consists of two layers. The first layer can find the optimum number of past time windows, and the second layer implements long-term prediction based on the obtained optimum number of windows. A series of experiments using publicly time series are conducted to assess the performance of the proposed architecture. The experimental results have revealed that the architecture has better performance in accuracy and stability.