A fuzzy interval time series energy and financial forecasting model using network-based multiple time-frequency spaces and the induced ordered weighted averaging aggregation operation

Forecasting time series is an emerging topic in operational research. Existing time series models have limited prediction accuracy when faced with the characteristics of nonlinearity and nonstationarity in complex situations related to energy and finance. To enhance overall prediction capabilities and improve forecasting accuracy, we propose a fuzzy interval time series forecasting model on the basis of network-based multiple time-frequency spaces and the induced ordered weighted averaging aggregation (IOWA) operation. Specifically, a time series signal is decomposed into ensemble empirical modes and then reconstructed as various time-frequency spaces, which are transformed into visibility graphs. Then, forecasting intervals in different spaces can be collected after the local random walker link prediction model is adopted. Furthermore, a rule-based representation value function inspired by Yager's golden rule approach is defined, and an appropriate representation value is calculated. Finally, after IOWA is used to aggregate the forecasting outcomes in different time-frequency spaces, the final forecast value can be obtained from the fuzzy forecasting interval. Considering that energy issues are of widespread interest in nature and the social economy, two cases, based on a hydrological time series from the Biliuhe River in China and two well-known sets of financial time series data, TAIEX and HSI, are studied to test the performance of the proposed approach in comparison with existing models. Our results show that the proposed approach can achieve better performance than well-developed models.