Time Series Prediction Using an Interval Arithmetic FIR Network

We present an interval arithmetic neural network based on the FIR (Finite Impulse Response) network to improve the prediction power and flexibility. The model for the system is obtained by modifying the dynamics of the FIR network to incorporate the interval arithmetic operations. In this paper, we describe the network architecture and behavior of the interval arithmetic FIR network. We present a learning method which is a modified temporal BP(Back Propagation) algorithm. The main attraction of this model is that the input data as well as the output data of the network can be represented and processed in the form of intervals. We claim that the proposed model is a generalized model of the FIR network with the capability of interval arithmetic operations. The effectiveness of the proposed model is evaluated through the experimental results of the long-range forecast for regional precipitation in Korea. Keywords—Interval arithmetic neural network, FIR network, time series prediction, and weather forecast