Article Per Year (5 Year)
p-Index From 2021 - 2026
0.23
P-Index
This Author published in this journals
All Journal International Journal of Power Electronics and Drive Systems (IJPEDS)
Are Sambasiva Rao
GITAM Deemed to be University
Control & Systems Engineering Electrical & Electronics Engineering

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

 1 
Adaptive forget-gated BiLSTM enhanced by DTW based feature selection in solar PV forecasting Are Sambasiva Rao; Kunada Dhana Sree Devi
International Journal of Power Electronics and Drive Systems (IJPEDS) Vol 17, No 3: September 2026
Publisher : Institute of Advanced Engineering and Science

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.11591/ijpeds.v17.i3.pp2086-2100

Abstract

Growing exhaustion of fuel reserves and their harmful environmental impacts have driven the shift towards the maintenance of renewable energy sources like solar energy. PV systems, however, still face significant challenges while integrating renewable sources into existing utility systems. Particularly, environmental features like temperature, irradiance, and humidity are not perfectly well coordinated with the power output, are always asynchronous in nature, and affect the power prediction considerably. Close observation of energy datasets from many PV plants revealed many intrinsic environmental variables that are highly asynchronous. Many forecasting models learn redundant features which might seem useful, and thereby the test performance is overfitting. To address the nonlinear and asynchronous behavior of environmental variables, there is a serious requirement for intelligent feature selection algorithms guided by both correlation and temporal alignment metrics. This research proposes a novel adaptive dynamic time warping (A-DWT) feature selection with an adaptive forget gate (AFG-BiLSTM) to address the asynchronous issues. Experiments were conducted with varied environmental asynchronous features, and the results of the proposed model were compared with traditional BiLSTM and stacked BiLSTM models. In all the experiments, the proposed model showed decreased error by (94.9%) on Dataset-1 (0.069, 0.0035), by (94.6%) on Dataset-2 (0.078, 0.0042), and by (90.9%) on Dataset-3 (0.069, 0.0061) when compared to stacked BiLSTM. The MSE loss of the proposed method was observed to be between (0.3%) and (11%) on four datasets.
Co-Authors Kunada Dhana Sree Devi