Time Series Forecasting

Time series is an important type of sequential data, and the underlying systems that generate these sequential data are often nonlinear dynamic systems. Therefore, most time series exhibit nonlinear and non-stationary characteristics. Nonlinear spiking neural P systems (in short, NSNP systems) are nonlinear neural-like computing models that can show rich dynamics, and are particularly suitable for processing sequential data. This chapter discusses how to use NSNP systems to achieve time series forecasting tasks, especially time series with nonlinear and non-stationary characteristics. Thus, three prediction models based on NSNP systems are discussed in detail, where the first is a univariate prediction model, the second is a multivariate prediction model, and the third is a recurrent-like prediction model. In the first two prediction models, redundant wavelet transform and non-subsampled shearlet transform are used as multi-scale time-frequency signal analysis tools.