A solar irradiance forecasting model using iterative filtering and bidirectional long short-term memory

ABSTRACT

Ensuring the effective functioning of solar-operated power systems requires an accurate solar irradiance forecasting. It increases the efficiency of planning and operation while ensuring the long-term viability of these systems. Consequently, a hybrid forecasting model for solar global horizontal irradiance is presented in this study utilizing a deep learning network called Bidirectional Long Short-Term Memory in conjunction with iterative filtering (IF). The function of IF is to split the historical time series data into several intrinsic mode functions. Furthermore, the preprocessing methodology: partial autocorrelation function is involved in the work to extract the relevant time lags as input features. Additionally, the grid search technique is utilized for the selection of the deep learning networks’ hyperparameters. The proposed model effectively compared with standalone models: Long Short-Term Memory, Gated Recurrent Unit, Bidirectional Long Short-Term Memory as well as recently developed complete ensemble mode decomposition with adaptive noise-based hybrid model. Taking different seasons of Gangtok, India location as a case, the proposed model achieved lower root mean square error (6.873 W/m² − 11.775 W/m²), mean absolute error (3.376 W/m² − 6.192 W/m²) and correlation coefficient (0.996–0.998) compared to contrast models (CEEMDAN-BILSTM, BILSTM, LSTM and GRU) for 15 minutes ahead forecast. For validation point of view, additional analysis is also carried out on the percentage improvement in results, multi-horizon forecasts (15, 30, 45, and 60 minutes ahead), and behavior on various day types.

KEYWORDS:

• Long short-term memory
• solar forecasting
• hybrid models
• bidirectional long short-term memory
• grid search
• complete ensemble empirical mode decomposition with adaptive noise
• iterative filtering

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data will be made available on request.

Additional information

Funding

This research is not supported by any funding agency.