Adaptive Broad Echo State Network for Nonstationary Time Series Forecasting

Time series forecasting provides a vital basis for the control and management of various systems. The time data in real world are usually strongly nonstationary nonlinear, which increases difficulty reliable forecasting. To fully utilize learning capability machine forecasting, an adaptive broad echo state network (ABESN) is proposed this paper. Firstly, system (BLS) used as framework, reservoir pools (ESN) introduced to form (BESN). Secondly, problem information redundancy structure BESN, optimization algorithm BESN based on pruning proposed. Thirdly, hyperparameters test index In brief, hyperparameter algorithms studied ABESN model applied air humidity electric load. experiments show that has better ability can achieve higher accuracy.