Multi-step ahead wind speed forecasting approach coupling PSR, NNCT-based multi-model fusion and a new optimization algorithm

Abstract

Wind-based electricity generation infrastructure continues to demonstrate substantial expansion rates in recent years. Such growth trajectories demand proportional evolution in wind power administration methodologies. Precise predictions represent an indispensable element for effective wind energy system governance. However, the task of generating accurate wind velocity forecasts remains challenging, since wind speed time-series data exhibits both non-linear patterns and temporal variability. This paper presents a novel hybrid model for wind speed forecasting that integrates PSR (Phase Space Reconstruction), NNCT (No Negative Constraint Theory), and an innovative GPSOGA optimization algorithm. SSA (Singular Spectrum Analysis) is initially applied to decompose the raw wind speed time series into IMFs (Intrinsic Mode Functions), effectively isolating fundamental oscillatory components. Subsequently, PSR reconstructs these IMFs into input and output vectors. The proposed model combines four predictive frameworks: CBP (Cascade Back Propagation) network, RNN (Recurrent Neural Network), GRU (Gated Recurrent Unit), and CCNRNN (Causal Convolutional Network integrated with Recurrent Neural Network). The NNCT strategy is employed to consolidate the outputs of these predictors, while a newly developed optimization algorithm identifies the optimal combination parameters. To evaluate the effectiveness of the proposed model, forecasting results are benchmarked against various models across four distinct datasets. Experimental results indicate that the proposed model achieves superior forecasting accuracy, as evidenced by multiple performance indicators. Further validation through the DM (Diebold-Mariano) test, AIC (Akaike's Information Criterion), and the NSE (Nash-Sutcliffe Efficiency Coefficient) confirms the model's enhanced predictive capability over comparison models.